feat(langgraph): add Runtime.on_exit for cleanup that runs however a run ends

release(langgraph): 1.2.19 (#9504 )
Bumps langgraph from 1.2.18 to 1.2.19. Includes: - List a `ToolNode`'s tool names in its `get_graph()` node metadata (#9449) - Apply a node's `trace_policy` to tracers only, so it no longer changes what streams see (#9486) - Give `messages` and `tools` events the namespace of the graph they ran in (#9458) - Stream messages a node returns from a graph streamed inside another node (#9463) - Respect `subgraphs=False` in `RemoteGraph.stream` when the parent streams subgraphs (#9453) - Stop reporting a tool paused on `interrupt()` as failed on the `tools` stream (#9406) - Return an awaited `@task`'s value while streaming instead of `None` (#9407)
2026-10-31 11:21:17 +01:00 · 2026-10-30 23:50:41 +00:00 · 2026-10-30 18:14:31 -04:00 · 2026-10-30 18:14:27 -04:00 · 2026-10-30 18:14:22 -04:00 · 2026-10-30 20:58:19 +00:00
84 changed files with 4960 additions and 846 deletions
@@ -24,6 +24,39 @@ TEST=path/to/test.py make test

 Other pytest arguments can also be supplied inside the `TEST` variable.

+## Direction: LangGraph is a runtime
+
+LangGraph is consolidating as the low-level runtime for agents and workflows:
+graph execution, checkpointing, streaming, and interrupt/resume. Critical
+production systems and agent harnesses run on it, so correctness and stability
+come first. It is moving away from being a friendly SDK for building agents
+directly; agent development belongs one layer up, on a harness — Deep Agents
+(github.com/langchain-ai/deepagents, a separate repo) is the recommended one.
+
+Optimize contributions for runtime correctness, reliability, performance, and
+simplicity — not for the shortest agent example.
+
+- One way to do each thing. A change never adds a way; removals are always
+  deliberate (see below). The number of ways only goes down. No aliases,
+  overloads, extra accepted types, implicit defaults, or customization hooks
+  for convenience: ergonomics alone never justify new public surface.
+- The mandate applies to everything that implements runtime behavior, in
+  whichever repo it lives: the graph engine (`langgraph`, `langgraph-core`),
+  the checkpointers (`checkpoint*`), and the server (`langgraph-api`). The
+  client SDKs and CLI stay ergonomic but are thin, faithful clients, never a
+  second implementation of runtime behavior.
+- Agent-level packages (`prebuilt`, `langgraph-supervisor`, `langgraph-swarm`,
+  `langgraph-cua`) are maintenance-only: fix bugs, don't add agent features.
+  New examples and docs that build agents should use Deep Agents.
+- Remove deliberately, not opportunistically. Removing or changing public
+  behavior must be the stated purpose of the PR and go through the deprecation
+  process — replacement in place, removal version recorded, migration note.
+  Never drop a default or overload as a side effect of an unrelated change.
+- Behavior changes land in Python and JS together (the implementations live in
+  separate repos; note in the PR when the other one needs a matching change).
+- If a task asks for something this section rules out, say so and propose the
+  runtime-level or harness-level alternative instead of quietly implementing it.
+
 ## Libraries

 The repository contains several Python and JavaScript/TypeScript libraries.
@@ -36,11 +36,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.4.22"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/25/ee/6caf7a40c36a1220410afe15a1cc64993a1f864871f698c0f93acb72842a/certifi-2026.4.22.tar.gz", hash = "sha256:8d455352a37b71bf76a79caa83a3d6c25afee4a385d632127b6afb3963f1c580", size = 137077, upload-time = "2026-04-22T11:26:11.191Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/22/30/7cd8fdcdfbc5b869528b079bfb76dcdf6056b1a2097a662e5e8c04f42965/certifi-2026.4.22-py3-none-any.whl", hash = "sha256:3cb2210c8f88ba2318d29b0388d1023c8492ff72ecdde4ebdaddbb13a31b1c4a", size = 135707, upload-time = "2026-04-22T11:26:09.372Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -279,7 +279,7 @@ wheels = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
    { name = "langchain-core" },
@@ -40,11 +40,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.1.4"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e0/2d/a891ca51311197f6ad14a7ef42e2399f36cf2f9bd44752b3dc4eab60fdc5/certifi-2026.1.4.tar.gz", hash = "sha256:ac726dd470482006e014ad384921ed6438c457018f4b3d204aea4281258b2120", size = 154268, upload-time = "2026-01-04T02:42:41.825Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e6/ad/3cc14f097111b4de0040c83a525973216457bbeeb63739ef1ed275c1c021/certifi-2026.1.4-py3-none-any.whl", hash = "sha256:9943707519e4add1115f44c2bc244f782c0249876bf51b6599fee1ffbedd685c", size = 152900, upload-time = "2026-01-04T02:42:40.15Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -278,7 +278,7 @@ wheels = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
    { name = "langchain-core" },
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"

 [project]
 name = "langgraph-checkpoint-sqlite"
-version = "3.1.2"
+version = "3.1.3"
 description = "Library with a SQLite implementation of LangGraph checkpoint saver."
 authors = []
 requires-python = ">=3.10"
@@ -12,7 +12,7 @@ readme = "README.md"
 license = "MIT"
 license-files = ['LICENSE']
 dependencies = [
-    "langgraph-checkpoint>=4.3.0,<5.0.0",
+    "langgraph-checkpoint>=4.3.1,<5.0.0",
    "aiosqlite>=0.20",
    "sqlite-vec>=0.1.6",
 ]
@@ -1364,6 +1364,17 @@ def test_search_empty_prefix_is_unconstrained(store: SqliteStore) -> None:
    }


+@pytest.mark.parametrize("batched", [False, True])
+def test_search_rejects_nul_namespace_prefix(store: SqliteStore, batched: bool) -> None:
+    store.put(("a",), "key", {"v": 1})
+
+    with pytest.raises(InvalidNamespaceError, match="NUL"):
+        if batched:
+            store.batch([SearchOp(("a\x00",))])
+        else:
+            store.search(("a\x00",))
+
+
 def test_namespace_labels_with_trailing_newline(store: SqliteStore) -> None:
    """Labels may contain newlines, and must not match a differently-named label.

@@ -49,11 +49,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.1.4"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e0/2d/a891ca51311197f6ad14a7ef42e2399f36cf2f9bd44752b3dc4eab60fdc5/certifi-2026.1.4.tar.gz", hash = "sha256:ac726dd470482006e014ad384921ed6438c457018f4b3d204aea4281258b2120", size = 154268, upload-time = "2026-01-04T02:42:41.825Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e6/ad/3cc14f097111b4de0040c83a525973216457bbeeb63739ef1ed275c1c021/certifi-2026.1.4-py3-none-any.whl", hash = "sha256:9943707519e4add1115f44c2bc244f782c0249876bf51b6599fee1ffbedd685c", size = 152900, upload-time = "2026-01-04T02:42:40.15Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -285,7 +285,7 @@ wheels = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
    { name = "langchain-core" },
@@ -360,7 +360,7 @@ test = [

 [[package]]
 name = "langgraph-checkpoint-sqlite"
-version = "3.1.2"
+version = "3.1.3"
 source = { editable = "." }
 dependencies = [
    { name = "aiosqlite" },
@@ -1333,6 +1333,10 @@ def _validate_namespace_labels(namespace: tuple[str, ...]) -> None:
            raise InvalidNamespaceError(
                f"Invalid namespace label '{label}' found in {namespace}. Namespace labels cannot contain periods ('.')."
            )
+        elif "\x00" in label:
+            raise InvalidNamespaceError(
+                f"Invalid namespace label '{label}' found in {namespace}. Namespace labels cannot contain NUL bytes ('\\x00')."
+            )
        elif not label:
            raise InvalidNamespaceError(
                f"Namespace labels cannot be empty strings. Got {label} in {namespace}"
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"

 [project]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 description = "Library with base interfaces for LangGraph checkpoint savers."
 authors = []
 requires-python = ">=3.10"
@@ -532,7 +532,7 @@ async def test_cannot_put_empty_namespace() -> None:
    assert (await async_store.aget(("valid", "namespace"), "key")) is None


-INVALID_NAMESPACES = [("foo.bar",), ("foo", ""), (123,)]
+INVALID_NAMESPACES = [("foo.bar",), ("foo\x00",), ("foo", ""), (123,)]
 NAMESPACE_METHODS = ["get", "delete", "search", "prefix", "suffix"]


@@ -51,11 +51,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.1.4"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e0/2d/a891ca51311197f6ad14a7ef42e2399f36cf2f9bd44752b3dc4eab60fdc5/certifi-2026.1.4.tar.gz", hash = "sha256:ac726dd470482006e014ad384921ed6438c457018f4b3d204aea4281258b2120", size = 154268, upload-time = "2026-01-04T02:42:41.825Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e6/ad/3cc14f097111b4de0040c83a525973216457bbeeb63739ef1ed275c1c021/certifi-2026.1.4-py3-none-any.whl", hash = "sha256:9943707519e4add1115f44c2bc244f782c0249876bf51b6599fee1ffbedd685c", size = 152900, upload-time = "2026-01-04T02:42:40.15Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -301,7 +301,7 @@ wheels = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "." }
 dependencies = [
    { name = "langchain-core" },
@@ -23,6 +23,7 @@ from langgraph_cli.schemas import (
    GraphDef,
    HttpConfig,
    IndexConfig,
+    PriorityConfig,
    SecurityConfig,
    SerdeConfig,
    StoreConfig,
@@ -113,6 +114,7 @@ def add_descriptions_to_schema(schema, cls):
                StoreConfig,
                IndexConfig,
                AuthConfig,
+                PriorityConfig,
                SecurityConfig,
                HttpConfig,
                CorsConfig,
@@ -1776,17 +1776,17 @@ baseline-browser-mapping@^2.11.12:
  integrity sha512-H0ulySigv6icDJ1F7SjtdCD6PrhTpdYCmP0CactWy1+ekh0AFd0o1Wn5T8b+hnTmdBx19u9yhL6wvCylXMY7zw==

 brace-expansion@^1.1.7:
-  version "1.1.11"
-  resolved "https://registry.yarnpkg.com/brace-expansion/-/brace-expansion-1.1.11.tgz#3c7fcbf529d87226f3d2f52b966ff5271eb441dd"
-  integrity sha512-iCuPHDFgrHX7H2vEI/5xpz07zSHB00TpugqhmYtVmMO6518mCuRMoOYFldEBl0g187ufozdaHgWKcYFb61qGiA==
+  version "1.1.21"
+  resolved "https://registry.yarnpkg.com/brace-expansion/-/brace-expansion-1.1.21.tgz#edf4fab5c64d051aea5a8def49aba1c7522279f3"
+  integrity sha512-9zeA+KLZNNzglF2TPKRQEDyx6Yby7daAkuy8MiPzpXPsYDWi/DRM8jmwUDxokQjYqBpv5DgPiwD4h4ZZSy1Ujw==
  dependencies:
    balanced-match "^1.0.0"
    concat-map "0.0.1"

 brace-expansion@^2.0.2:
-  version "2.0.2"
-  resolved "https://registry.yarnpkg.com/brace-expansion/-/brace-expansion-2.0.2.tgz#54fc53237a613d854c7bd37463aad17df87214e7"
-  integrity sha512-Jt0vHyM+jmUBqojB7E1NIYadt0vI0Qxjxd2TErW94wDz+E2LAm5vKMXXwg6ZZBTHPuUlDgQHKXvjGBdfcF1ZDQ==
+  version "2.1.7"
+  resolved "https://registry.yarnpkg.com/brace-expansion/-/brace-expansion-2.1.7.tgz#14ea3836731823bb24740225ea43b941b436f9d1"
+  integrity sha512-uZbew1NqdmPDTMJ8ah1y+b+9QEJrfkXFk3RcTQw3X0jW/xRUvFKsg1CfQdSYGdTbXZWExtU3J3ccxtnfw1Fi0g==
  dependencies:
    balanced-match "^1.0.0"

@@ -5217,6 +5217,6 @@ yocto-queue@^0.1.0:
  integrity sha512-rVksvsnNCdJ/ohGc6xgPwyN8eheCxsiLM8mxuE/t/mOVqJewPuO1miLpTHQiRgTKCLexL4MeAFVagts7HmNZ2Q==

 "zod@^3.25.76 || ^4":
-  version "4.3.6"
-  resolved "https://registry.yarnpkg.com/zod/-/zod-4.3.6.tgz#89c56e0aa7d2b05107d894412227087885ab112a"
-  integrity sha512-rftlrkhHZOcjDwkGlnUtZZkvaPHCsDATp4pGpuOOMDaTdDDXF91wuVDJoWoPsKX/3YPQ5fHuF3STjcYyKr+Qhg==
+  version "4.6.7"
+  resolved "https://registry.yarnpkg.com/zod/-/zod-4.6.7.tgz#e806cf64a7f4e37635ce9f2f52c04c3766f75e5b"
+  integrity sha512-g8W2D2kU6a8untc053fX1uVBKzM7yqAxW/3zwk83+6RyF2+SKOuV+c0wBDhUwzV1PaGapmZH+rmcUzPf+3CGwg==
@@ -581,9 +581,9 @@ base64-js@^1.5.1:
  integrity sha512-AKpaYlHn8t4SVbOHCy+b5+KKgvR4vrsD8vbvrbiQJps7fKDTkjkDry6ji0rUJjC0kzbNePLwzxq8iypo41qeWA==

 brace-expansion@^1.1.7:
-  version "1.1.12"
-  resolved "https://registry.yarnpkg.com/brace-expansion/-/brace-expansion-1.1.12.tgz#ab9b454466e5a8cc3a187beaad580412a9c5b843"
-  integrity sha512-9T9UjW3r0UW5c1Q7GTwllptXwhvYmEzFhzMfZ9H7FQWt+uZePjZPjBP/W1ZEyZ1twGWom5/56TF4lPcqjnDHcg==
+  version "1.1.21"
+  resolved "https://registry.yarnpkg.com/brace-expansion/-/brace-expansion-1.1.21.tgz#edf4fab5c64d051aea5a8def49aba1c7522279f3"
+  integrity sha512-9zeA+KLZNNzglF2TPKRQEDyx6Yby7daAkuy8MiPzpXPsYDWi/DRM8jmwUDxokQjYqBpv5DgPiwD4h4ZZSy1Ujw==
  dependencies:
    balanced-match "^1.0.0"
    concat-map "0.0.1"
@@ -2148,6 +2148,6 @@ yocto-queue@^0.1.0:
  integrity sha512-rVksvsnNCdJ/ohGc6xgPwyN8eheCxsiLM8mxuE/t/mOVqJewPuO1miLpTHQiRgTKCLexL4MeAFVagts7HmNZ2Q==

 "zod@^3.25.76 || ^4":
-  version "4.3.6"
-  resolved "https://registry.yarnpkg.com/zod/-/zod-4.3.6.tgz#89c56e0aa7d2b05107d894412227087885ab112a"
-  integrity sha512-rftlrkhHZOcjDwkGlnUtZZkvaPHCsDATp4pGpuOOMDaTdDDXF91wuVDJoWoPsKX/3YPQ5fHuF3STjcYyKr+Qhg==
+  version "4.6.7"
+  resolved "https://registry.yarnpkg.com/zod/-/zod-4.6.7.tgz#e806cf64a7f4e37635ce9f2f52c04c3766f75e5b"
+  integrity sha512-g8W2D2kU6a8untc053fX1uVBKzM7yqAxW/3zwk83+6RyF2+SKOuV+c0wBDhUwzV1PaGapmZH+rmcUzPf+3CGwg==
@@ -1 +1 @@
-__version__ = "0.4.37"
+__version__ = "0.4.40"
@@ -847,6 +847,7 @@ def dev(
        store=config_json.get("store"),
        wait_for_client=wait_for_client,
        auth=config_json.get("auth"),
+        priority=config_json.get("priority"),
        http=config_json.get("http"),
        ui=config_json.get("ui"),
        ui_config=config_json.get("ui_config"),
@@ -442,6 +442,7 @@ def validate_config(
        "store": config.get("store"),
        "auth": config.get("auth"),
        "encryption": config.get("encryption"),
+        "priority": config.get("priority"),
        "http": config.get("http"),
        # Pass through webhooks config so it can be injected into the image
        "webhooks": config.get("webhooks"),
@@ -606,6 +607,19 @@ def validate_config(
                    f"Invalid encryption.path format: '{encryption_conf['path']}'. "
                    "Must be in format './path/to/file.py:attribute_name'"
                )
+    # Validate priority config
+    if priority_conf := config.get("priority"):
+        if "path" in priority_conf:
+            if ":" not in priority_conf["path"]:
+                raise ValueError(
+                    f"Invalid priority.path format: '{priority_conf['path']}'. "
+                    "Must be in format './path/to/file.py:attribute_name'"
+                )
+            if _is_node_graph(priority_conf["path"]):
+                raise ValueError(
+                    "JavaScript priority callbacks are not supported: "
+                    f"'{priority_conf['path']}'. Use a Python file or module instead."
+                )
    if http_conf := config.get("http"):
        if "app" in http_conf:
            if ":" not in http_conf["app"]:
@@ -649,6 +663,7 @@ _KNOWN_CONFIG_KEYS = {
    "store",
    "auth",
    "encryption",
+    "priority",
    "http",
    "webhooks",
    "checkpointer",
@@ -1080,6 +1095,49 @@ def _update_encryption_path(
    )


+def _update_priority_path(
+    config_path: pathlib.Path, config: Config, local_deps: LocalDeps
+) -> None:
+    """Update priority.path to use Docker container paths."""
+    priority_conf = config.get("priority")
+    if not priority_conf or not (path_str := priority_conf.get("path")):
+        return
+
+    module_str, sep, attr_str = path_str.partition(":")
+    if not sep or not module_str.startswith("."):
+        return  # Already validated or absolute path
+
+    resolved = config_path.parent / module_str
+    if not resolved.exists():
+        raise FileNotFoundError(
+            f"Priority file not found: {resolved} (from {path_str})"
+        )
+    if not resolved.is_file():
+        raise IsADirectoryError(f"Priority path must be a file: {resolved}")
+
+    # Check faux packages first (they take precedence)
+    for faux_path, (_, destpath) in local_deps.faux_pkgs.items():
+        if resolved.is_relative_to(faux_path):
+            new_path = f"{destpath}/{resolved.relative_to(faux_path)}:{attr_str}"
+            priority_conf["path"] = new_path
+            return
+
+    # Check real packages
+    for real_path in local_deps.real_pkgs:
+        if resolved.is_relative_to(real_path):
+            new_path = (
+                f"/deps/{real_path.name}/{resolved.relative_to(real_path)}:{attr_str}"
+            )
+            priority_conf["path"] = new_path
+            return
+
+    raise ValueError(
+        f"Priority file '{resolved}' not covered by dependencies.\n"
+        "Add its parent directory to the 'dependencies' array in your config.\n"
+        f"Current dependencies: {config['dependencies']}"
+    )
+
+
 def _update_checkpointer_path(
    config_path: pathlib.Path, config: Config, local_deps: LocalDeps
 ) -> None:
@@ -1213,6 +1271,9 @@ def _build_runtime_env_vars(config: Config) -> list[str]:
    if (encryption_config := config.get("encryption")) is not None:
        env_vars.append(f"ENV LANGGRAPH_ENCRYPTION='{json.dumps(encryption_config)}'")

+    if (priority_config := config.get("priority")) is not None:
+        env_vars.append(f"ENV LANGGRAPH_PRIORITY='{json.dumps(priority_config)}'")
+
    if (http_config := config.get("http")) is not None:
        env_vars.append(f"ENV LANGGRAPH_HTTP='{json.dumps(http_config)}'")

@@ -1380,6 +1441,8 @@ def python_config_to_docker(
    _update_auth_path(config_path, config, local_deps)
    # Rewrite encryption path, so it points to the correct location in the Docker container
    _update_encryption_path(config_path, config, local_deps)
+    # Rewrite priority path, so it points to the correct location in the Docker container
+    _update_priority_path(config_path, config, local_deps)
    # Rewrite checkpointer path, so it points to the correct location in the Docker container
    _update_checkpointer_path(config_path, config, local_deps)
    # Rewrite HTTP app path, so it points to the correct location in the Docker container
@@ -12,7 +12,7 @@ from collections.abc import Callable, Mapping, Sequence
 from contextlib import contextmanager
 from dataclasses import asdict, dataclass, field
 from datetime import datetime, timezone
-from typing import Protocol, TypeVar
+from typing import Generic, Protocol, TypeVar

 import click
 import click.exceptions
@@ -26,7 +26,6 @@ from langgraph_cli.dependency_tracking import find_tracked_packages
 from langgraph_cli.docker import build_docker_image, can_build_locally
 from langgraph_cli.exec import CommandRunner, Runner, subp_exec
 from langgraph_cli.host_backend import (
-    MAX_PAGE_SIZE,
    ControlPlaneEndpoints,
    HostBackendClient,
    HostBackendError,
@@ -183,12 +182,15 @@ class Listener:
        )
        return cls(
            identifier,
-            str(resource.get("compute_id", "")),
+            str(resource.get("compute_id") or ""),
            tuple(str(namespace) for namespace in namespaces)
            if isinstance(namespaces, list)
            else (),
        )

+    def cells(self) -> tuple[str | None, ...]:
+        return (self.id, self.compute_id, ", ".join(self.namespaces))
+

@dataclass(frozen=True, slots=True)
 class Unplaced:
@@ -276,8 +278,6 @@ def _describe_listeners(listeners: Sequence[Listener]) -> str:
    ]
    if len(listeners) > len(shown):
        lines.append(f"  ... and {len(listeners) - len(shown)} more")
-    if len(listeners) == MAX_PAGE_SIZE:
-        lines.append(f"  (only the first {MAX_PAGE_SIZE} listeners were read)")
    return "\n".join(lines)


@@ -499,16 +499,9 @@ def _source_of(resource: object) -> str | None:
 def find_deployment_by_name(
    client: HostBackendClient, name: str
 ) -> ExistingDeployment | None:
-    listed = client.list_deployments(name=name, name_contains=name, limit=MAX_PAGE_SIZE)
-    for resource in listed:
+    for resource in client.list_deployments(name=name, name_contains=name):
        if resource.get("name") == name and resource.get("id"):
            return ExistingDeployment(str(resource["id"]), _source_of(resource))
-    if len(listed) >= MAX_PAGE_SIZE:
-        raise click.ClickException(
-            "This workspace has more deployments than the CLI can search, so it "
-            f"cannot tell whether '{name}' already exists. Pass --deployment-id to "
-            "update an existing deployment."
-        )
    return None


@@ -583,103 +576,96 @@ def _validate_prebuilt_image(
    _get_emitter().info(f"Image is available for {_DEPLOYMENT_PLATFORM}")


-def _extract_deployment_url(deployment: dict[str, object]) -> str:
-    source_config = deployment.get("source_config")
-    if isinstance(source_config, dict):
-        custom_url = source_config.get("custom_url")
-        if isinstance(custom_url, str) and custom_url:
-            return custom_url
-    return "-"
+Resource = Mapping[str, object]
+RowT = TypeVar("RowT")


-def format_deployments_table(deployments: Sequence[dict[str, object]]) -> str:
-    headers = ("Deployment ID", "Deployment Name", "Deployment URL")
-    rows = [
-        (
-            str(deployment.get("id", "-") or "-"),
-            str(deployment.get("name", "-") or "-"),
-            _extract_deployment_url(deployment),
-        )
-        for deployment in deployments
-    ]
-    widths = [
-        max(len(headers[index]), *(len(row[index]) for row in rows))
-        for index in range(len(headers))
-    ]
+@dataclass(frozen=True, slots=True)
+class Table(Generic[RowT]):
+    headers: tuple[str, ...]
+    cells: Callable[[RowT], tuple[str | None, ...]]

-    def format_row(row: Sequence[str]) -> str:
-        return "  ".join(value.ljust(widths[index]) for index, value in enumerate(row))
-
-    lines = [format_row(headers), format_row(tuple("-" * width for width in widths))]
-    lines.extend(format_row(row) for row in rows)
-    return "\n".join(lines)
-
-
-def _extract_listener_namespaces(listener: dict[str, object]) -> str:
-    compute_config = listener.get("compute_config")
-    namespaces = (
-        compute_config.get("k8s_namespaces")
-        if isinstance(compute_config, dict)
-        else None
-    )
-    if isinstance(namespaces, list) and namespaces:
-        return ", ".join(str(namespace) for namespace in namespaces)
-    return "-"
-
-
-def format_listeners_table(listeners: Sequence[dict[str, object]]) -> str:
-    headers = ("Listener ID", "Compute ID", "Namespaces")
-    rows = [
-        (
-            str(listener.get("id", "-") or "-"),
-            str(listener.get("compute_id", "-") or "-"),
-            _extract_listener_namespaces(listener),
-        )
-        for listener in listeners
-    ]
-    widths = [
-        max(len(headers[index]), *(len(row[index]) for row in rows))
-        for index in range(len(headers))
-    ]
-
-    def format_row(row: Sequence[str]) -> str:
-        return "  ".join(value.ljust(widths[index]) for index, value in enumerate(row))
-
-    lines = [format_row(headers), format_row(tuple("-" * width for width in widths))]
-    lines.extend(format_row(row) for row in rows)
-    return "\n".join(lines)
-
-
-def format_revisions_table(revisions: Sequence[dict[str, object]]) -> str:
-    headers = ("Revision ID", "Status", "Created At")
-    latest_deployed_seen = False
-    rows = []
-    for revision in revisions:
-        status = str(revision.get("status", "-") or "-")
-        if status == "DEPLOYED":
-            if latest_deployed_seen:
-                status = "REPLACED"
-            else:
-                latest_deployed_seen = True
-        rows.append(
-            (
-                str(revision.get("id", "-") or "-"),
-                status,
-                str(revision.get("created_at", "-") or "-"),
+    def render(self, rows: Sequence[RowT]) -> str:
+        grid = [
+            self.headers,
+            *(tuple(cell or "-" for cell in self.cells(row)) for row in rows),
+        ]
+        widths = [max(map(len, column)) for column in zip(*grid, strict=True)]
+        rule = tuple("-" * width for width in widths)
+        return "\n".join(
+            "  ".join(
+                cell.ljust(width) for cell, width in zip(line, widths, strict=True)
            )
+            for line in (grid[0], rule, *grid[1:])
        )

-    widths = [
-        max(len(headers[index]), *(len(row[index]) for row in rows))
-        for index in range(len(headers))
+
+def _text(resource: Resource, key: str) -> str | None:
+    value = resource.get(key)
+    return str(value) if value else None
+
+
+def _deployment_url(deployment: Resource) -> str | None:
+    source_config = deployment.get("source_config")
+    custom_url = (
+        source_config.get("custom_url") if isinstance(source_config, Mapping) else None
+    )
+    return custom_url if isinstance(custom_url, str) and custom_url else None
+
+
+def _with_replaced_status(revisions: Sequence[Resource]) -> list[Resource]:
+    live_index = next(
+        (
+            index
+            for index, revision in enumerate(revisions)
+            if revision.get("status") == "DEPLOYED"
+        ),
+        None,
+    )
+    return [
+        {**revision, "status": "REPLACED"}
+        if revision.get("status") == "DEPLOYED" and index != live_index
+        else revision
+        for index, revision in enumerate(revisions)
    ]

-    def format_row(row: Sequence[str]) -> str:
-        return "  ".join(value.ljust(widths[index]) for index, value in enumerate(row))

-    lines = [format_row(headers), format_row(tuple("-" * width for width in widths))]
-    lines.extend(format_row(row) for row in rows)
-    return "\n".join(lines)
+@dataclass(frozen=True, slots=True)
+class DeploymentRow:
+    id: str
+    name: str | None
+    url: str | None
+
+    @classmethod
+    def from_resource(cls, deployment: Resource) -> "DeploymentRow":
+        identifier = _text(deployment, "id")
+        if identifier is None:
+            raise HostBackendError(
+                "The control plane returned a deployment without an id."
+            )
+        return cls(identifier, _text(deployment, "name"), _deployment_url(deployment))
+
+    def cells(self) -> tuple[str | None, ...]:
+        return (self.id, self.name, self.url)
+
+    def as_json(self) -> dict[str, str | None]:
+        return {"id": self.id, "name": self.name}
+
+
+DEPLOYMENTS_TABLE = Table(
+    ("Deployment ID", "Deployment Name", "Deployment URL"), DeploymentRow.cells
+)
+
+LISTENERS_TABLE = Table(("Listener ID", "Compute ID", "Namespaces"), Listener.cells)
+
+REVISIONS_TABLE = Table(
+    ("Revision ID", "Status", "Created At"),
+    lambda revision: (
+        _text(revision, "id"),
+        _text(revision, "status"),
+        _text(revision, "created_at"),
+    ),
+)


 def format_timestamp(ts) -> str:
@@ -841,7 +827,6 @@ def _find_deployment(
            lambda c: c.list_deployments(
                agent_id=selector.agent_id,
                agent_environment=selector.environment,
-                limit=MAX_PAGE_SIZE,
            ),
        )
        if len(existing) > 1:
@@ -1007,20 +992,15 @@ def _print_deployment_result(
    dashboard_label: str,
    status_url: str | None = None,
 ) -> None:
-    """Print final deployment status and raise on failure."""
    em = _get_emitter()
-    dep_info = client.get_deployment(deployment_id)
-    custom_url = None
-    if isinstance(dep_info, dict):
-        sc = dep_info.get("source_config")
-        if isinstance(sc, dict):
-            custom_url = sc.get("custom_url")
+    deployment = client.get_deployment(deployment_id)
+    url = _deployment_url(deployment) if isinstance(deployment, Mapping) else None

    if last_status == "DEPLOYED":
        em.result(
            "succeeded",
            deployment_id=deployment_id,
-            url=custom_url,
+            url=url,
            status_url=status_url,
        )
    elif last_status in ("BUILD_FAILED", "DEPLOY_FAILED", "CREATE_FAILED"):
@@ -1081,6 +1061,8 @@ def _docker_config_for_token(registry_host: str, token: str, username: str):

 _UPLOAD_TIMEOUT_SECONDS = 300
 _BYTES_PER_MIB = 1_048_576
+# Byte range Cloud Storage signs into upload URLs, up to the 200 MB archive limit.
+_GCS_SIZE_RANGE = f"0,{200 * _BYTES_PER_MIB}"


 class _ProgressReader:
@@ -1106,12 +1088,24 @@ class _ProgressReader:
        return self._file_size


-def _upload_to_gcs(signed_url: str, file_path: str, file_size: int) -> None:
-    """Upload tarball to GCS via signed PUT URL with progress display."""
+def _upload_source_tarball(
+    signed_url: str,
+    file_path: str,
+    file_size: int,
+    upload_headers: dict[str, str] | None,
+) -> None:
+    """Upload tarball via signed PUT URL with progress display."""
    import urllib.error
    import urllib.request

    em = _get_emitter()
+    # Control planes that return no `upload_headers` sign for the Cloud Storage
+    # size range. Content-Type is a fallback: without one, urllib sends a form
+    # content type, which breaks signatures that cover it.
+    headers = {
+        "Content-Type": "application/gzip",
+        **(upload_headers or {"X-Goog-Content-Length-Range": _GCS_SIZE_RANGE}),
+    }

    with open(file_path, "rb") as f:
        reader = _ProgressReader(f, file_size, em)
@@ -1119,11 +1113,7 @@ def _upload_to_gcs(signed_url: str, file_path: str, file_size: int) -> None:
            signed_url,
            data=reader,
            method="PUT",
-            headers={
-                "Content-Type": "application/gzip",
-                "Content-Length": str(file_size),
-                "X-Goog-Content-Length-Range": "0,209715200",
-            },
+            headers={**headers, "Content-Length": str(file_size)},
        )
        try:
            urllib.request.urlopen(req, timeout=_UPLOAD_TIMEOUT_SECONDS)
@@ -1443,7 +1433,9 @@ def _run_remote_build(
        step += 1

        _log_deploy_step(step, "Uploading source")
-        _upload_to_gcs(signed_url, archive_path, file_size)
+        _upload_source_tarball(
+            signed_url, archive_path, file_size, upload_data.get("upload_headers")
+        )
    step += 1

    _log_deploy_step(step, "Triggering remote build")
@@ -2594,6 +2586,13 @@ def _deploy_cmd(
    default="",
    help="Only show deployments whose names contain this value.",
 )
+@click.option(
+    "--json",
+    "json_output",
+    is_flag=True,
+    default=False,
+    help="Print the deployments as a JSON array of {id, name} instead of a table.",
+)
@deploy.command("list", help="[Beta] List LangSmith Deployments.")
 def deploy_list(
    api_key: str | None,
@@ -2601,11 +2600,13 @@ def deploy_list(
    name_contains: str,
    agent_id: str | None,
    environment: str | None,
+    json_output: bool,
 ) -> None:
    if agent_id is not None or environment is not None:
        click.secho(
            "Note: --agent-id and --agent-environment flags are in private beta",
            fg="yellow",
+            err=True,
        )
    if agent_id is not None and not agent_id.strip():
        raise click.UsageError("--agent-id must not be empty.")
@@ -2614,15 +2615,19 @@ def deploy_list(
        filters["agent_id"] = agent_id
    if environment is not None:
        filters["agent_environment"] = environment
-    client = _create_host_backend_client(host_url, api_key)
-    deployments = _call_host_backend_with_optional_tenant(
+    client = _create_host_backend_client(host_url, api_key, no_input=json_output)
+    resources = _call_host_backend_with_optional_tenant(
        client,
        lambda c: c.list_deployments(name_contains=name_contains, **filters),
+        no_input=json_output,
    )
-    if not deployments:
+    deployments = [DeploymentRow.from_resource(resource) for resource in resources]
+    if json_output:
+        click.echo(json_mod.dumps([deployment.as_json() for deployment in deployments]))
+    elif not deployments:
        click.echo("No deployments found.")
-        return
-    click.echo(format_deployments_table(deployments))
+    else:
+        click.echo(DEPLOYMENTS_TABLE.render(deployments))


 # ---------------------------------------------------------------------------
@@ -2665,11 +2670,14 @@ def deploy_listeners_list(
    )
    if json_output:
        click.echo(json_mod.dumps(listeners))
-        return
-    if not listeners:
+    elif not listeners:
        click.echo("No listeners found for this workspace.")
-        return
-    click.echo(format_listeners_table(listeners))
+    else:
+        click.echo(
+            LISTENERS_TABLE.render(
+                [Listener.from_resource(listener) for listener in listeners]
+            )
+        )


 # ---------------------------------------------------------------------------
@@ -2712,7 +2720,7 @@ def deploy_revisions_list(
    if not revisions:
        click.echo(f"No revisions found for deployment {deployment_id}.")
        return
-    click.echo(format_revisions_table(revisions))
+    click.echo(REVISIONS_TABLE.render(_with_replaced_status(revisions)))


 # ---------------------------------------------------------------------------
@@ -2,6 +2,7 @@

 from __future__ import annotations

+from collections.abc import Mapping
 from dataclasses import dataclass
 from typing import Any, Literal
 from urllib.parse import urlparse
@@ -220,27 +221,51 @@ class HostBackendClient:
        *,
        name: str | None = None,
        name_contains: str | None = None,
-        limit: int | None = None,
        agent_id: str | None = None,
        agent_environment: str | None = None,
-    ) -> list[dict[str, Any]]:
+    ) -> list[dict[str, object]]:
        given = (
            ("name", name),
            ("name_contains", name_contains),
-            ("limit", limit),
            ("agent_id", agent_id),
            ("agent_environment", agent_environment),
        )
-        params = {key: value for key, value in given if value is not None}
-        return _resources(self._request("GET", "/v2/deployments", params=params))
+        filters = {key: value for key, value in given if value is not None}
+        return self._every_page("/v2/deployments", filters)
+
+    def _every_page(
+        self, path: str, filters: Mapping[str, object]
+    ) -> list[dict[str, object]]:
+        resources: list[dict[str, object]] = []
+        seen_ids: set[object] = set()
+        while True:
+            page = _resources(
+                self._request(
+                    "GET",
+                    path,
+                    params={
+                        **filters,
+                        "limit": MAX_PAGE_SIZE,
+                        "offset": len(resources),
+                    },
+                )
+            )
+            page_ids = {resource.get("id") for resource in page}
+            if page_ids and page_ids <= seen_ids:
+                raise HostBackendError(
+                    f"GET {path} returned the same resources again: this control "
+                    "plane ignores 'offset', so the CLI cannot read past the first page."
+                )
+            seen_ids |= page_ids
+            resources.extend(page)
+            if len(page) < MAX_PAGE_SIZE:
+                return resources

    def get_listener(self, listener_id: str) -> dict[str, Any]:
        return self._request("GET", f"/v2/listeners/{listener_id}")

-    def list_listeners(self) -> list[dict[str, Any]]:
-        return _resources(
-            self._request("GET", "/v2/listeners", params={"limit": MAX_PAGE_SIZE})
-        )
+    def list_listeners(self) -> list[dict[str, object]]:
+        return self._every_page("/v2/listeners", {})

    def get_deployment(self, deployment_id: str) -> dict[str, Any]:
        return self._request("GET", f"/v2/deployments/{deployment_id}")
@@ -258,10 +283,13 @@ class HostBackendClient:
        )

    def request_upload_url(self, deployment_id: str) -> dict[str, Any]:
-        """Get a signed GCS URL for uploading the source tarball."""
+        """Get a signed URL for uploading the source tarball."""
+        # Tells the control plane this client sends the returned
+        # `upload_headers` with its upload, which some storage requires.
        return self._request(
            "POST",
            f"/v2/deployments/{deployment_id}/upload-url",
+            {"supports_upload_headers": True},
        )

    def update_deployment(
@@ -373,6 +373,27 @@ class EncryptionConfig(TypedDict, total=False):
    """


+class PriorityConfig(TypedDict, total=False):
+    """Configuration for custom run priority logic.
+
+    Allows you to decide each thread's priority settings with a Python function
+    that the server calls at the first run on a thread with no settings, which
+    can come long after the thread was created."""
+
+    path: str
+    """Required. Path to an instance of the Priority() class that registers an assign function.
+
+    Format: "path/to/file.py:my_priority"
+
+    Example:
+        {
+            "priority": {
+                "path": "./priority.py:my_priority"
+            }
+        }
+    """
+
+
 class CorsConfig(TypedDict, total=False):
    """Specifies Cross-Origin Resource Sharing (CORS) rules for your server.

@@ -753,6 +774,13 @@ class Config(TypedDict, total=False):
    Allows you to implement custom encryption for sensitive data stored in the database.
    """

+    priority: PriorityConfig | None
+    """Optional. Custom run priority config, including the path to your Python priority logic.
+
+    Allows you to decide each thread's priority settings at the first run on a thread with no settings,
+    which can come long after the thread was created.
+    """
+
    http: HttpConfig | None
    """Optional. Configuration for the built-in HTTP server, controlling which custom routes are exposed
    and how cross-origin requests are handled.
@@ -784,6 +812,7 @@ __all__ = [
    "CheckpointerConfig",
    "AuthConfig",
    "EncryptionConfig",
+    "PriorityConfig",
    "HttpConfig",
    "MiddlewareOrders",
    "Distros",
@@ -910,6 +910,7 @@ def python_config_to_docker_uv_lock(
    for section, key in [
        ("auth", "path"),
        ("encryption", "path"),
+        ("priority", "path"),
        ("checkpointer", "path"),
        ("http", "app"),
    ]:
@@ -196,6 +196,17 @@
                }
              ]
            },
+            "priority": {
+              "anyOf": [
+                {
+                  "$ref": "#/$defs/PriorityConfig"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "description": "Optional. Custom run priority config, including the path to your Python priority logic.\n\nAllows you to decide each thread's priority settings at the first run on a thread with no settings,\nwhich can come long after the thread was created.\n"
+            },
            "source": {
              "anyOf": [
                {
@@ -438,6 +449,17 @@
                }
              ]
            },
+            "priority": {
+              "anyOf": [
+                {
+                  "$ref": "#/$defs/PriorityConfig"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "description": "Optional. Custom run priority config, including the path to your Python priority logic.\n\nAllows you to decide each thread's priority settings at the first run on a thread with no settings,\nwhich can come long after the thread was created.\n"
+            },
            "source": {
              "$ref": "#/$defs/UvSource"
            },
@@ -666,6 +688,17 @@
                }
              ]
            },
+            "priority": {
+              "anyOf": [
+                {
+                  "$ref": "#/$defs/PriorityConfig"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "description": "Optional. Custom run priority config, including the path to your Python priority logic.\n\nAllows you to decide each thread's priority settings at the first run on a thread with no settings,\nwhich can come long after the thread was created.\n"
+            },
            "source": {
              "anyOf": [
                {
@@ -1170,6 +1203,18 @@
      },
      "required": []
    },
+    "PriorityConfig": {
+      "title": "PriorityConfig",
+      "description": "Configuration for custom run priority logic.\n\nAllows you to decide each thread's priority settings with a Python function\nthat the server calls at the first run on a thread with no settings, which\ncan come long after the thread was created.",
+      "type": "object",
+      "properties": {
+        "path": {
+          "type": "string",
+          "description": "Required. Path to an instance of the Priority() class that registers an assign function.\n"
+        }
+      },
+      "required": []
+    },
    "UvSource": {
      "title": "UvSource",
      "description": "Deployment source rooted at a uv project or workspace.",
@@ -196,6 +196,17 @@
                }
              ]
            },
+            "priority": {
+              "anyOf": [
+                {
+                  "$ref": "#/$defs/PriorityConfig"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "description": "Optional. Custom run priority config, including the path to your Python priority logic.\n\nAllows you to decide each thread's priority settings at the first run on a thread with no settings,\nwhich can come long after the thread was created.\n"
+            },
            "source": {
              "anyOf": [
                {
@@ -438,6 +449,17 @@
                }
              ]
            },
+            "priority": {
+              "anyOf": [
+                {
+                  "$ref": "#/$defs/PriorityConfig"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "description": "Optional. Custom run priority config, including the path to your Python priority logic.\n\nAllows you to decide each thread's priority settings at the first run on a thread with no settings,\nwhich can come long after the thread was created.\n"
+            },
            "source": {
              "$ref": "#/$defs/UvSource"
            },
@@ -666,6 +688,17 @@
                }
              ]
            },
+            "priority": {
+              "anyOf": [
+                {
+                  "$ref": "#/$defs/PriorityConfig"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "description": "Optional. Custom run priority config, including the path to your Python priority logic.\n\nAllows you to decide each thread's priority settings at the first run on a thread with no settings,\nwhich can come long after the thread was created.\n"
+            },
            "source": {
              "anyOf": [
                {
@@ -1170,6 +1203,18 @@
      },
      "required": []
    },
+    "PriorityConfig": {
+      "title": "PriorityConfig",
+      "description": "Configuration for custom run priority logic.\n\nAllows you to decide each thread's priority settings with a Python function\nthat the server calls at the first run on a thread with no settings, which\ncan come long after the thread was created.",
+      "type": "object",
+      "properties": {
+        "path": {
+          "type": "string",
+          "description": "Required. Path to an instance of the Priority() class that registers an assign function.\n"
+        }
+      },
+      "required": []
+    },
    "UvSource": {
      "title": "UvSource",
      "description": "Deployment source rooted at a uv project or workspace.",
@@ -2,10 +2,12 @@ import json
 import pathlib
 import re
 import shutil
+import sys
 import tempfile
 import textwrap
 from contextlib import contextmanager
 from pathlib import Path
+from unittest.mock import MagicMock

 import click
 from click.testing import CliRunner
@@ -371,6 +373,32 @@ def test_dev_command_requires_ssl_certfile_and_keyfile_together(tmp_path) -> Non
    assert "Both --ssl-certfile and --ssl-keyfile must be provided" in result.output


+def test_dev_command_passes_priority_to_the_server(tmp_path, monkeypatch) -> None:
+    config_path = tmp_path / "langgraph.json"
+    config_path.write_text(
+        json.dumps(
+            {
+                "dependencies": ["."],
+                "graphs": {"agent": "./agent.py:graph"},
+                "priority": {"path": "./priority.py:priority"},
+            }
+        ),
+        encoding="utf-8",
+    )
+    # The test environment doesn't install langgraph-api, so stand in for its server.
+    server = MagicMock()
+    monkeypatch.setitem(sys.modules, "langgraph_api.cli", server)
+    # dev appends the working directory and its dependencies to sys.path.
+    monkeypatch.setattr(sys, "path", list(sys.path))
+
+    result = CliRunner().invoke(cli, ["dev", "--config", str(config_path)])
+
+    assert result.exit_code == 0, result.output
+    assert server.run_server.call_args.kwargs["priority"] == {
+        "path": "./priority.py:priority"
+    }
+
+
 def test_deploy_list_command(monkeypatch) -> None:
    runner = CliRunner()
    captured: dict[str, str] = {}
@@ -454,6 +482,177 @@ def test_deploy_list_command_no_results(monkeypatch) -> None:
    assert result.output.strip() == "No deployments found."


+def test_deploy_list_prints_an_aligned_table(monkeypatch) -> None:
+    class FakeClient:
+        def __init__(self, host_url: str, api_key: str, tenant_id: str | None = None):
+            pass
+
+        def list_deployments(self, name_contains: str = ""):
+            return [
+                {
+                    "id": "dep-123",
+                    "name": "alpha",
+                    "source_config": {"custom_url": "https://alpha.example.com"},
+                },
+                {"id": "dep-4567", "name": "beta-agent", "source_config": {}},
+            ]
+
+    monkeypatch.setattr(deploy_module, "HostBackendClient", FakeClient)
+
+    result = CliRunner().invoke(
+        cli,
+        [
+            "deploy",
+            "list",
+            "--api-key",
+            "test-key",
+            "--host-url",
+            "https://api.example.com",
+        ],
+    )
+
+    assert result.exit_code == 0, result.output
+    assert result.output.splitlines() == [
+        "Deployment ID  Deployment Name  Deployment URL           ",
+        "-------------  ---------------  -------------------------",
+        "dep-123        alpha            https://alpha.example.com",
+        "dep-4567       beta-agent       -                        ",
+    ]
+
+
+def _deploy_list_json(
+    monkeypatch, deployments: list[dict], *extra_args: str
+) -> click.testing.Result:
+    class FakeClient:
+        def __init__(self, host_url: str, api_key: str, tenant_id: str | None = None):
+            pass
+
+        def list_deployments(
+            self,
+            name_contains: str = "",
+            agent_id: str | None = None,
+            agent_environment: str | None = None,
+        ):
+            return deployments
+
+    monkeypatch.setattr(deploy_module, "HostBackendClient", FakeClient)
+    return CliRunner().invoke(
+        cli,
+        [
+            "deploy",
+            "list",
+            "--json",
+            "--api-key",
+            "test-key",
+            "--host-url",
+            "https://api.example.com",
+            *extra_args,
+        ],
+    )
+
+
+def test_deploy_list_command_json(monkeypatch) -> None:
+    result = _deploy_list_json(
+        monkeypatch,
+        [
+            {
+                "id": "dep-123",
+                "name": "alpha",
+                "source_config": {"custom_url": "https://alpha.example.com"},
+            },
+            {"id": "dep-456", "name": "beta", "source_config": {}},
+        ],
+    )
+
+    assert result.exit_code == 0, result.output
+    assert json.loads(result.stdout) == [
+        {"id": "dep-123", "name": "alpha"},
+        {"id": "dep-456", "name": "beta"},
+    ]
+
+
+def test_deploy_list_json_never_prints_deployment_secrets(monkeypatch) -> None:
+    result = _deploy_list_json(
+        monkeypatch,
+        [
+            {
+                "id": "dep-123",
+                "name": "alpha",
+                "source_config": {"custom_url": None},
+                "secrets": [{"name": "OPENAI_API_KEY", "value": "sk-plain-value"}],
+            }
+        ],
+    )
+
+    assert result.exit_code == 0, result.output
+    assert "sk-plain-value" not in result.output
+    assert json.loads(result.stdout) == [{"id": "dep-123", "name": "alpha"}]
+
+
+def test_deploy_list_command_json_no_results(monkeypatch) -> None:
+    result = _deploy_list_json(monkeypatch, [])
+
+    assert result.exit_code == 0, result.output
+    assert json.loads(result.stdout) == []
+
+
+def test_deploy_list_rejects_a_deployment_without_an_id(monkeypatch) -> None:
+    result = _deploy_list_json(monkeypatch, [{"name": "alpha"}])
+
+    assert result.exit_code != 0
+    assert "deployment without an id" in result.output
+
+
+def test_deploy_list_json_keeps_the_agent_note_off_stdout(monkeypatch) -> None:
+    result = _deploy_list_json(
+        monkeypatch,
+        [{"id": "dep-123", "name": "alpha"}],
+        "--agent-id",
+        "customer-support",
+        "--agent-environment",
+        "staging",
+    )
+
+    assert result.exit_code == 0, result.output
+    assert json.loads(result.stdout) == [{"id": "dep-123", "name": "alpha"}]
+    assert "private beta" in result.stderr
+
+
+def test_deploy_list_json_never_prompts_for_a_workspace(monkeypatch) -> None:
+    class FakeClient:
+        def __init__(self, host_url: str, api_key: str, tenant_id: str | None = None):
+            pass
+
+        def list_deployments(self, name_contains: str = ""):
+            raise HostBackendError("requires workspace specification", status_code=403)
+
+        def set_tenant(self, tenant_id: str) -> None:
+            raise AssertionError("must not retry with a prompted tenant with --json")
+
+    monkeypatch.setattr(deploy_module, "HostBackendClient", FakeClient)
+    prompt_calls = []
+    monkeypatch.setattr(
+        click, "prompt", lambda *a, **k: prompt_calls.append((a, k)) or "unreachable"
+    )
+
+    result = CliRunner().invoke(
+        cli,
+        [
+            "deploy",
+            "list",
+            "--json",
+            "--api-key",
+            "test-key",
+            "--host-url",
+            "https://api.example.com",
+        ],
+    )
+
+    assert prompt_calls == []
+    assert result.exit_code != 0
+    assert "LANGSMITH_TENANT_ID" in result.output
+
+
 def test_deploy_listeners_list_command(monkeypatch) -> None:
    runner = CliRunner()
    captured: dict[str, str] = {}
@@ -536,6 +735,84 @@ def test_deploy_listeners_list_command_no_results(monkeypatch) -> None:
    assert result.output.strip() == "No listeners found for this workspace."


+def test_deploy_listeners_list_prints_an_aligned_table(monkeypatch) -> None:
+    class FakeClient:
+        def __init__(self, host_url: str, api_key: str, tenant_id: str | None = None):
+            pass
+
+        def list_listeners(self):
+            return [
+                {
+                    "id": "listener-1",
+                    "compute_id": "prod-cluster",
+                    "compute_config": {"k8s_namespaces": ["agents"]},
+                },
+                {
+                    "id": "listener-22",
+                    "compute_id": "multi",
+                    "compute_config": {"k8s_namespaces": ["agents", "agents-staging"]},
+                },
+                {"id": "listener-3", "compute_id": "", "compute_config": {}},
+                {
+                    "id": "listener-4",
+                    "compute_id": None,
+                    "compute_config": {"k8s_namespaces": ["agents"]},
+                },
+            ]
+
+    monkeypatch.setattr(deploy_module, "HostBackendClient", FakeClient)
+
+    result = CliRunner().invoke(
+        cli,
+        [
+            "deploy",
+            "listeners",
+            "list",
+            "--api-key",
+            "test-key",
+            "--host-url",
+            "https://api.example.com",
+        ],
+    )
+
+    assert result.exit_code == 0, result.output
+    assert result.output.splitlines() == [
+        "Listener ID  Compute ID    Namespaces            ",
+        "-----------  ------------  ----------------------",
+        "listener-1   prod-cluster  agents                ",
+        "listener-22  multi         agents, agents-staging",
+        "listener-3   -             -                     ",
+        "listener-4   -             agents                ",
+    ]
+
+
+def test_deploy_listeners_list_rejects_a_listener_without_an_id(monkeypatch) -> None:
+    class FakeClient:
+        def __init__(self, host_url: str, api_key: str, tenant_id: str | None = None):
+            pass
+
+        def list_listeners(self):
+            return [{"compute_id": "prod-cluster"}]
+
+    monkeypatch.setattr(deploy_module, "HostBackendClient", FakeClient)
+
+    result = CliRunner().invoke(
+        cli,
+        [
+            "deploy",
+            "listeners",
+            "list",
+            "--api-key",
+            "test-key",
+            "--host-url",
+            "https://api.example.com",
+        ],
+    )
+
+    assert result.exit_code != 0
+    assert "listener without an id" in result.output
+
+
 def test_deploy_listeners_list_command_json(monkeypatch) -> None:
    runner = CliRunner()

@@ -738,6 +1015,54 @@ def test_deploy_revisions_list_command_no_results(monkeypatch) -> None:
    assert result.output.strip() == "No revisions found for deployment dep-123."


+def test_deploy_revisions_list_marks_older_deployed_revisions_as_replaced(
+    monkeypatch,
+) -> None:
+    class FakeClient:
+        def __init__(self, host_url: str, api_key: str, tenant_id: str | None = None):
+            pass
+
+        def list_revisions(self, deployment_id: str, limit: int = 1):
+            return [
+                {
+                    "id": "rev-3",
+                    "status": "DEPLOYING",
+                    "created_at": "2026-10-20T10:00:00Z",
+                },
+                {
+                    "id": "rev-2",
+                    "status": "DEPLOYED",
+                    "created_at": "2026-10-19T10:00:00Z",
+                },
+                {"id": "rev-1", "status": "DEPLOYED", "created_at": None},
+            ]
+
+    monkeypatch.setattr(deploy_module, "HostBackendClient", FakeClient)
+
+    result = CliRunner().invoke(
+        cli,
+        [
+            "deploy",
+            "revisions",
+            "list",
+            "--api-key",
+            "test-key",
+            "--host-url",
+            "https://api.example.com",
+            "dep-123",
+        ],
+    )
+
+    assert result.exit_code == 0, result.output
+    assert result.output.splitlines() == [
+        "Revision ID  Status     Created At          ",
+        "-----------  ---------  --------------------",
+        "rev-3        DEPLOYING  2026-10-20T10:00:00Z",
+        "rev-2        DEPLOYED   2026-10-19T10:00:00Z",
+        "rev-1        REPLACED   -                   ",
+    ]
+
+
 def test_deploy_revisions_list_command_with_explicit_limit(monkeypatch) -> None:
    runner = CliRunner()
    captured: dict[str, str] = {}
@@ -96,6 +96,7 @@ class ControlPlaneDouble:
    existing_deployments: list[dict] = field(default_factory=list)
    push_token_status: int = 200
    push_token_extra: dict = field(default_factory=dict)
+    upload_url_extra: dict = field(default_factory=dict)
    create_error: str | None = None
    listeners: list[dict] = field(default_factory=list)
    listeners_by_id: dict[str, dict] = field(default_factory=dict)
@@ -111,7 +112,11 @@ class ControlPlaneDouble:
    def _respond(self, request: httpx.Request) -> httpx.Response:
        method, path = request.method, request.url.path
        if (method, path) == ("GET", "/v2/listeners"):
-            return httpx.Response(200, json={"resources": self.listeners})
+            offset = int(request.url.params.get("offset", 0))
+            limit = int(request.url.params.get("limit", 20))
+            return httpx.Response(
+                200, json={"resources": self.listeners[offset : offset + limit]}
+            )
        if method == "GET" and path.startswith("/v2/listeners/"):
            listener_id = path.rsplit("/", 1)[-1]
            if not _looks_like_a_uuid(listener_id):
@@ -159,7 +164,12 @@ class ControlPlaneDouble:
            )
        if path.endswith("/upload-url"):
            return httpx.Response(
-                200, json={"upload_url": SIGNED_UPLOAD_URL, "object_path": OBJECT_PATH}
+                200,
+                json={
+                    "upload_url": SIGNED_UPLOAD_URL,
+                    "object_path": OBJECT_PATH,
+                    **self.upload_url_extra,
+                },
            )
        if method == "PATCH":
            return httpx.Response(200, json={"tenant_id": "tenant-1"})
@@ -276,7 +286,7 @@ class DeployProject:
    control_plane: ControlPlaneDouble
    docker: DockerDouble
    timeline: list[str]
-    uploads: list[tuple[str, str, int]]
+    uploads: list[tuple[str, str, int, dict[str, str] | None]]

    def run(self, *args: str, host_url: str = CONTROL_PLANE_URL) -> Result:
        return CliRunner().invoke(
@@ -313,16 +323,21 @@ def deploy_project(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> DeployPro
    timeline: list[str] = []
    control_plane = ControlPlaneDouble(timeline)
    docker = DockerDouble(timeline)
-    uploads: list[tuple[str, str, int]] = []
+    uploads: list[tuple[str, str, int, dict[str, str] | None]] = []

    @contextmanager
    def fake_create_archive(config_path: Path, config: dict) -> Iterator[tuple]:
        timeline.append("create_archive")
        yield ARCHIVE

-    def fake_upload(signed_url: str, file_path: str, file_size: int) -> None:
+    def fake_upload(
+        signed_url: str,
+        file_path: str,
+        file_size: int,
+        upload_headers: dict[str, str] | None,
+    ) -> None:
        timeline.append("upload_archive")
-        uploads.append((signed_url, file_path, file_size))
+        uploads.append((signed_url, file_path, file_size, upload_headers))

    monkeypatch.setattr(deploy_module, "_no_input", False)
    monkeypatch.setattr(deploy_module, "_emitter", None)
@@ -340,7 +355,7 @@ def deploy_project(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> DeployPro
    )
    monkeypatch.setattr(deploy_module.platform, "machine", lambda: "x86_64")
    monkeypatch.setattr(archive_module, "create_archive", fake_create_archive)
-    monkeypatch.setattr(deploy_module, "_upload_to_gcs", fake_upload)
+    monkeypatch.setattr(deploy_module, "_upload_source_tarball", fake_upload)
    return DeployProject(control_plane, docker, timeline, uploads)


@@ -699,6 +714,9 @@ def test_deployment_not_created_by_the_cli_gets_an_actionable_error(
 def test_remote_build_creates_an_internal_source_deployment_and_uploads_the_archive(
    deploy_project: DeployProject,
 ) -> None:
+    upload_headers = {"Content-Type": "application/gzip", "x-ms-blob-type": "BlockBlob"}
+    deploy_project.control_plane.upload_url_extra = {"upload_headers": upload_headers}
+
    result = deploy_project.run("--remote", "--install-command", "yarn install")

    assert result.exit_code == 0, result.output
@@ -713,7 +731,12 @@ def test_remote_build_creates_an_internal_source_deployment_and_uploads_the_arch
    assert deploy_project.control_plane.bodies[CREATE_DEPLOYMENT]["source"] == (
        "internal_source"
    )
-    assert deploy_project.uploads == [(SIGNED_UPLOAD_URL, ARCHIVE[0], ARCHIVE[1])]
+    assert deploy_project.control_plane.bodies[_upload_url(CREATED_ID)] == {
+        "supports_upload_headers": True
+    }
+    assert deploy_project.uploads == [
+        (SIGNED_UPLOAD_URL, ARCHIVE[0], ARCHIVE[1], upload_headers)
+    ]
    assert deploy_project.control_plane.bodies[_patch(CREATED_ID)] == {
        "revision_source": "internal_source",
        "source_revision_config": {
@@ -1288,14 +1311,16 @@ def test_listener_flags_without_push_to_make_no_call_at_all(
    assert deploy_project.timeline == []


-def test_a_truncated_listener_page_says_so(deploy_project: DeployProject) -> None:
+def test_listeners_beyond_the_first_page_are_counted(
+    deploy_project: DeployProject,
+) -> None:
    deploy_project.control_plane.listeners = [
        {
            "id": str(uuid.UUID(int=index)),
            "compute_id": "cluster",
            "compute_config": {"k8s_namespaces": ["agents"]},
        }
-        for index in range(100)
+        for index in range(101)
    ]

    result = deploy_project.run(
@@ -1303,7 +1328,8 @@ def test_a_truncated_listener_page_says_so(deploy_project: DeployProject) -> Non
    )

    assert result.exit_code != 0
-    assert "first 100" in result.output
+    assert "... and 91 more" in result.output
+    assert "first 100" not in result.output


 def test_a_managed_build_in_a_listener_workspace_points_at_push_to(
@@ -16,6 +16,7 @@ from langgraph_cli.config import (
    config_to_docker,
    default_base_image,
    docker_tag,
+    get_unknown_keys,
    has_disallowed_build_command_content,
    validate_config,
    validate_config_file,
@@ -155,6 +156,7 @@ def test_validate_config():
        "store": None,
        "auth": None,
        "encryption": None,
+        "priority": None,
        "webhooks": None,
        "checkpointer": None,
        "http": None,
@@ -184,6 +186,7 @@ def test_validate_config():
        "store": None,
        "auth": None,
        "encryption": None,
+        "priority": None,
        "webhooks": None,
        "checkpointer": None,
        "http": None,
@@ -1289,6 +1292,109 @@ def test_config_to_docker_python_encryption_formatted():
    )


+@pytest.mark.parametrize(
+    "path", ["./priority.py:priority", "my_package.priority:priority"]
+)
+def test_config_to_docker_python_priority(path: str):
+    # Test that priority config is kept by validation and isn't an unknown key
+    raw_config = {
+        "python_version": "3.11",
+        "graphs": {"agent": "./agent.py:graph"},
+        "dependencies": ["."],
+        "priority": {"path": path},
+    }
+    validated = validate_config(raw_config)
+
+    assert validated["priority"] == {"path": path}
+    assert get_unknown_keys(raw_config) == []
+
+
+def test_config_to_docker_python_priority_bad_path():
+    # Test that invalid priority path format raises ValueError
+    with pytest.raises(ValueError, match="Invalid priority.path format"):
+        validate_config(
+            {
+                "python_version": "3.11",
+                "graphs": {"agent": "./agent.py:graph"},
+                "dependencies": ["."],
+                "priority": {"path": "./priority.py"},  # Missing :attribute
+            }
+        )
+
+
+@pytest.mark.parametrize("ext", [".ts", ".mts", ".cts", ".js", ".mjs", ".cjs"])
+def test_config_to_docker_python_priority_js_path(ext: str):
+    # Test that a JavaScript priority path is rejected
+    with pytest.raises(ValueError, match="JavaScript priority callbacks"):
+        validate_config(
+            {
+                "python_version": "3.11",
+                "graphs": {"agent": "./agent.py:graph"},
+                "dependencies": ["."],
+                "priority": {"path": f"./priority{ext}:priority"},
+            }
+        )
+
+
+def test_config_to_docker_nodejs_priority():
+    # Test that a Node.js-only project can name a Python priority file
+    actual_docker_stdin, _ = config_to_docker(
+        PATH_TO_CONFIG,
+        validate_config(
+            {
+                "node_version": "20",
+                "graphs": {"agent": "./graphs/agent.js:graph"},
+                "priority": {"path": "./priority.py:priority"},
+            }
+        ),
+        base_image="langchain/langgraphjs-api",
+    )
+    assert (
+        """ENV LANGGRAPH_PRIORITY='{"path": "./priority.py:priority"}'"""
+        in actual_docker_stdin
+    )
+
+
+def test_config_to_docker_python_priority_formatted():
+    # Test that priority config is written to the Docker output with its container path
+    actual_docker_stdin, _ = config_to_docker(
+        PATH_TO_CONFIG,
+        validate_config(
+            {
+                "python_version": "3.11",
+                "dependencies": ["."],
+                "graphs": {"agent": "./graphs/agent.py:graph"},
+                "priority": {"path": "./agent.py:my_priority"},
+            }
+        ),
+        base_image="langchain/langgraph-api",
+    )
+    assert (
+        "ENV LANGGRAPH_PRIORITY="
+        """'{"path": "/deps/outer-unit_tests/unit_tests/agent.py:my_priority"}'"""
+        in actual_docker_stdin
+    )
+
+
+def test_config_to_docker_python_priority_not_in_dependencies():
+    # Test that a priority file outside the dependencies raises ValueError
+    with pytest.raises(
+        ValueError, match="Priority file .* not covered by dependencies"
+    ):
+        config_to_docker(
+            PATH_TO_CONFIG,
+            validate_config(
+                {
+                    "python_version": "3.11",
+                    "dependencies": ["./graphs"],
+                    "graphs": {"agent": "./graphs/agent.py:graph"},
+                    "priority": {"path": "./agent.py:my_priority"},
+                }
+            ),
+            base_image="langchain/langgraph-api",
+        )
+
+
 def test_config_to_docker_nodejs_internal_docker_tag():
    graphs = {"agent": "./graphs/agent.js:graph"}
    actual_docker_stdin, additional_contexts = config_to_docker(
@@ -1731,6 +1837,38 @@ def test_config_to_docker_uv_lock_honors_root_workspace_sources():
        )


+def test_config_to_docker_uv_lock_rewrites_priority_path():
+    with tempfile.TemporaryDirectory() as tmpdir:
+        tmpdir_path = pathlib.Path(tmpdir)
+        project_root, config_path = _write_uv_lock_workspace(
+            tmpdir_path,
+            agent_sources="[tool.uv.sources]\nshared = { workspace = true }",
+        )
+        (project_root / "libs" / "shared" / "src" / "shared" / "priority.py").touch()
+
+        config = validate_config(
+            {
+                "python_version": "3.11",
+                "graphs": {
+                    "agent": "../../apps/agent/src/agent/graph.py:graph",
+                },
+                "source": {"kind": "uv", "root": "../..", "package": "agent"},
+                "priority": {
+                    "path": "../../libs/shared/src/shared/priority.py:priority"
+                },
+            }
+        )
+        docker, _ = config_to_docker(
+            config_path, config, base_image="langchain/langgraph-api:0.2.47"
+        )
+
+        assert (
+            "ENV LANGGRAPH_PRIORITY="
+            """'{"path": "/deps/workspace/libs/shared/src/shared/priority.py:priority"}'"""
+            in docker
+        )
+
+
 def test_config_to_docker_uv_lock_ignores_unrelated_workspace_package_sources():
    with tempfile.TemporaryDirectory() as tmpdir:
        tmpdir_path = pathlib.Path(tmpdir)
@@ -75,6 +75,7 @@ def test_agent_create(deployment_api, tmp_path, monkeypatch):
        "agent_id": "customer-support",
        "agent_environment": "staging",
        "limit": "100",
+        "offset": "0",
    }
    payload = json.loads(requests[1].content)
    assert payload["agent"] == {
@@ -33,6 +33,7 @@ from langgraph_cli.deploy import (
    _Emitter,
    _env_without_deployment_name,
    _parse_env_from_config,
+    _print_deployment_result,
    _resolve_env_path,
    _resolve_pushed_image_digest,
    _resource_spec,
@@ -536,6 +537,39 @@ class TestEmitterTextMode:
        assert "URL: https://app.test" in lines


+@pytest.mark.parametrize(
+    ("deployment", "expected_url_lines"),
+    [
+        pytest.param(
+            {
+                "id": "dep-1",
+                "source_config": {"custom_url": "https://agent.example.com"},
+            },
+            ["URL: https://agent.example.com"],
+            id="custom_url_is_printed",
+        ),
+        pytest.param(
+            {"id": "dep-1", "source_config": {}}, [], id="no_custom_url_no_line"
+        ),
+    ],
+)
+def test_a_deployed_result_prints_the_deployment_url(
+    capsys, monkeypatch, deployment, expected_url_lines
+):
+    class FakeClient:
+        def get_deployment(self, deployment_id: str) -> dict:
+            return deployment
+
+    monkeypatch.setattr(deploy_mod, "_emitter", None)
+
+    _print_deployment_result(
+        FakeClient(), "dep-1", "DEPLOYED", dashboard_label="Deployment dashboard"
+    )
+
+    lines = [line.strip() for line in capsys.readouterr().out.splitlines()]
+    assert [line for line in lines if line.startswith("URL:")] == expected_url_lines
+
+
 # ---------------------------------------------------------------------------
 # --no-input guard on _create_host_backend_client
 # ---------------------------------------------------------------------------
@@ -1210,6 +1244,7 @@ def test_finding_a_deployment_by_name_narrows_the_search_for_every_server_versio
        "name": "agent",
        "name_contains": "agent",
        "limit": "100",
+        "offset": "0",
    }
    assert found == ExistingDeployment("dep-1", "github")

@@ -1237,20 +1272,29 @@ def test_a_server_that_ignores_the_exact_name_filter_never_matches_another_deplo
    assert find_deployment_by_name(client, "brand-new-agent") is None


-def test_a_full_page_without_a_match_refuses_to_claim_the_name_is_free():
-    page = [
-        {"id": f"dep-{index}", "name": f"other-agent-{index}"} for index in range(100)
-    ]
-    client = HostBackendClient(
-        "https://api.example.com",
-        "key",
-        transport=httpx.MockTransport(
-            lambda req: httpx.Response(200, json={"resources": page})
+def test_a_deployment_beyond_the_first_page_is_found_by_name():
+    deployments = [
+        *(
+            {"id": f"dep-{index}", "name": f"other-agent-{index}"}
+            for index in range(100)
        ),
+        {"id": "dep-target", "name": "agent", "source": "external_docker"},
+    ]
+
+    def handler(req: httpx.Request) -> httpx.Response:
+        offset = int(req.url.params["offset"])
+        limit = int(req.url.params["limit"])
+        return httpx.Response(
+            200, json={"resources": deployments[offset : offset + limit]}
+        )
+
+    client = HostBackendClient(
+        "https://api.example.com", "key", transport=httpx.MockTransport(handler)
    )

-    with pytest.raises(click.ClickException, match="--deployment-id"):
-        find_deployment_by_name(client, "brand-new-agent")
+    assert find_deployment_by_name(client, "agent") == ExistingDeployment(
+        "dep-target", "external_docker"
+    )


 def test_a_partial_page_without_a_match_means_the_name_is_free():
@@ -1414,3 +1458,59 @@ class TestPlacementOnAKnownListener:
    def test_refuses_and_names_the_namespaces(self, request_, listener, message):
        with pytest.raises(click.UsageError, match=message):
            request_.on(listener)
+
+
+DEFAULT_UPLOAD_HEADERS = {
+    "content-type": "application/gzip",
+    "content-length": "7",
+    "x-goog-content-length-range": "0,209715200",
+}
+
+
+@pytest.mark.parametrize(
+    ("upload_headers", "expected"),
+    [
+        pytest.param(None, DEFAULT_UPLOAD_HEADERS, id="missing_sends_defaults"),
+        pytest.param({}, DEFAULT_UPLOAD_HEADERS, id="empty_sends_defaults"),
+        pytest.param(
+            {
+                "content-type": "application/x-gzip",
+                "x-ms-blob-type": "BlockBlob",
+                "content-length": "1",
+            },
+            {
+                "content-type": "application/x-gzip",
+                "x-ms-blob-type": "BlockBlob",
+                "content-length": "7",
+            },
+            id="returned_headers_replace_defaults",
+        ),
+        pytest.param(
+            {"x-ms-blob-type": "BlockBlob"},
+            {
+                "content-type": "application/gzip",
+                "x-ms-blob-type": "BlockBlob",
+                "content-length": "7",
+            },
+            id="returned_headers_keep_a_content_type",
+        ),
+    ],
+)
+def test_source_upload_sends_the_headers_the_signed_url_requires(
+    tmp_path, mocker, upload_headers, expected
+):
+    archive = tmp_path / "source.tar.gz"
+    archive.write_bytes(b"tarball")
+    sent = []
+    mocker.patch(
+        "langgraph_cli.deploy._get_emitter", return_value=MagicMock(json_mode=True)
+    )
+    mocker.patch(
+        "urllib.request.urlopen", side_effect=lambda req, timeout: sent.append(req)
+    )
+
+    deploy_mod._upload_source_tarball(
+        "https://storage.example/signed", str(archive), 7, upload_headers
+    )
+
+    assert {name.lower(): value for name, value in sent[0].header_items()} == expected
@@ -575,7 +575,7 @@ def test_list_listeners_asks_for_a_full_page():
    )

    assert c.list_listeners() == [{"id": "listener-1"}]
-    assert seen["url"] == "https://api.example.com/v2/listeners?limit=100"
+    assert seen["url"] == "https://api.example.com/v2/listeners?limit=100&offset=0"


@pytest.mark.parametrize(
@@ -602,29 +602,30 @@ def test_is_cloud_recognises_the_managed_control_plane(control_plane_url, expect
    assert endpoints.is_cloud is expected


+FIRST_PAGE = {"limit": "100", "offset": "0"}
+
+
@pytest.mark.parametrize(
    ("call", "expected_params"),
    [
        pytest.param(
            lambda c: c.list_deployments(name="agent"),
-            {"name": "agent"},
+            {"name": "agent", **FIRST_PAGE},
            id="exact_name_filters_server_side",
        ),
        pytest.param(
            lambda c: c.list_deployments(name_contains="age"),
-            {"name_contains": "age"},
+            {"name_contains": "age", **FIRST_PAGE},
            id="substring_search_keeps_its_own_parameter",
        ),
        pytest.param(
            lambda c: c.list_deployments(),
-            {},
-            id="no_filter_sends_no_parameters",
+            FIRST_PAGE,
+            id="no_filter_sends_only_the_page",
        ),
        pytest.param(
-            lambda c: c.list_deployments(
-                name="agent", name_contains="agent", limit=100
-            ),
-            {"name": "agent", "name_contains": "agent", "limit": "100"},
+            lambda c: c.list_deployments(name="agent", name_contains="agent"),
+            {"name": "agent", "name_contains": "agent", **FIRST_PAGE},
            id="both_filters_travel_together_for_older_servers",
        ),
    ],
@@ -645,6 +646,68 @@ def test_list_deployments_sends_one_name_filter(call, expected_params):
    assert seen == expected_params


+@pytest.mark.parametrize(
+    "list_every",
+    [
+        pytest.param(HostBackendClient.list_deployments, id="deployments"),
+        pytest.param(HostBackendClient.list_listeners, id="listeners"),
+    ],
+)
+def test_list_endpoints_read_every_page(list_every):
+    resources = [{"id": f"resource-{index}"} for index in range(101)]
+    offsets: list[str] = []
+
+    def handler(req: httpx.Request) -> httpx.Response:
+        offset = int(req.url.params["offset"])
+        limit = int(req.url.params["limit"])
+        offsets.append(req.url.params["offset"])
+        return httpx.Response(
+            200, json={"resources": resources[offset : offset + limit]}
+        )
+
+    c = HostBackendClient(
+        "https://api.example.com", "key", transport=httpx.MockTransport(handler)
+    )
+
+    assert list_every(c) == resources
+    assert offsets == ["0", "100"]
+
+
+FIRST_PAGE_OF_DEPLOYMENTS = [
+    {"id": f"dep-{index}", "status": "DEPLOYED"} for index in range(100)
+]
+
+
+@pytest.mark.parametrize(
+    "repeated_page",
+    [
+        pytest.param(FIRST_PAGE_OF_DEPLOYMENTS, id="same_page_again"),
+        pytest.param(
+            [
+                {"id": "dep-99", "status": "DEPLOYING"},
+                *FIRST_PAGE_OF_DEPLOYMENTS[:99],
+            ],
+            id="same_deployments_reordered_and_updated",
+        ),
+    ],
+)
+def test_list_deployments_fails_when_the_server_ignores_the_offset(repeated_page):
+    pages = iter([FIRST_PAGE_OF_DEPLOYMENTS, repeated_page])
+
+    def handler(req: httpx.Request) -> httpx.Response:
+        page = next(pages, None)
+        if page is None:
+            return httpx.Response(500, json={"detail": "the guard let paging run on"})
+        return httpx.Response(200, json={"resources": page})
+
+    c = HostBackendClient(
+        "https://api.example.com", "key", transport=httpx.MockTransport(handler)
+    )
+
+    with pytest.raises(HostBackendError, match="ignores 'offset'"):
+        c.list_deployments()
+
+
@pytest.mark.parametrize(
    ("body", "expected"),
    [
@@ -1,11 +1,6 @@
 from unittest.mock import patch

-from langgraph_cli.deploy import (
-    _extract_deployment_url,
-    format_deployments_table,
-    format_listeners_table,
-    format_revisions_table,
-)
+from langgraph_cli.deploy import Table
 from langgraph_cli.util import clean_empty_lines, warn_non_wolfi_distro


@@ -194,96 +189,7 @@ def test_warn_non_wolfi_distro_does_not_modify_config():
    assert config_copy == original_config  # Config should remain unchanged


-def test_extract_deployment_url_uses_custom_url():
-    deployment = {"source_config": {"custom_url": "https://example.com/custom"}}
-    assert _extract_deployment_url(deployment) == "https://example.com/custom"
+def test_a_table_without_rows_still_renders_its_headers():
+    table = Table(("Name", "Status"), lambda resource: (None, None))

-
-def test_extract_deployment_url_defaults_to_dash():
-    assert _extract_deployment_url({"id": "dep-123"}) == "-"
-
-
-def test_format_deployments_table():
-    output = format_deployments_table(
-        [
-            {
-                "id": "dep-123",
-                "name": "alpha",
-                "source_config": {"custom_url": "https://alpha.example.com"},
-            },
-            {
-                "id": "dep-456",
-                "name": "beta",
-                "url": "https://beta.example.com",
-            },
-        ]
-    )
-    assert "Deployment ID" in output
-    assert "Deployment Name" in output
-    assert "Deployment URL" in output
-    assert "dep-123" in output
-    assert "alpha" in output
-    assert "https://alpha.example.com" in output
-    assert "dep-456" in output
-
-
-def test_format_revisions_table():
-    output = format_revisions_table(
-        [
-            {
-                "id": "rev-123",
-                "status": "DEPLOYED",
-                "created_at": "2023-11-09T10:00:00Z",
-            },
-            {
-                "id": "rev-456",
-                "status": "CREATING",
-                "created_at": "2023-11-07T05:31:56Z",
-            },
-            {
-                "id": "rev-789",
-                "status": "DEPLOYED",
-                "created_at": "2023-11-08T10:00:00Z",
-            },
-        ]
-    )
-    assert "Revision ID" in output
-    assert "Status" in output
-    assert "Created At" in output
-    assert "rev-123" in output
-    assert "CREATING" in output
-    assert "2023-11-07T05:31:56Z" in output
-    assert "rev-456" in output
-    assert "rev-789" in output
-    assert "REPLACED" in output
-
-
-def test_format_listeners_table():
-    output = format_listeners_table(
-        [
-            {
-                "id": "listener-1",
-                "compute_id": "prod-cluster",
-                "compute_config": {"k8s_namespaces": ["agents"]},
-            },
-            {
-                "id": "listener-2",
-                "compute_id": "multi-cluster",
-                "compute_config": {"k8s_namespaces": ["agents", "agents-staging"]},
-            },
-            {
-                "id": "listener-3",
-                "compute_id": "broken-cluster",
-            },
-        ]
-    )
-    assert "Listener ID" in output
-    assert "Compute ID" in output
-    assert "Namespaces" in output
-    assert "listener-1" in output
-    assert "prod-cluster" in output
-    assert "agents" in output
-    assert "listener-2" in output
-    assert "agents, agents-staging" in output
-    assert "listener-3" in output
-    assert "broken-cluster" in output
+    assert table.render([]).splitlines() == ["Name  Status", "----  ------"]
@@ -1,5 +1,5 @@
 version = 1
-revision = 3
+revision = 5
 requires-python = ">=3.11"

 [manifest]
@@ -52,11 +52,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.2.25"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/af/2d/7bf41579a8986e348fa033a31cdd0e4121114f6bce2457e8876010b092dd/certifi-2026.2.25.tar.gz", hash = "sha256:e887ab5cee78ea814d3472169153c2d12cd43b14bd03329a39a9c6e2e80bfba7", size = 155029, upload-time = "2026-02-25T02:54:17.342Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/9a/3c/c17fb3ca2d9c3acff52e30b309f538586f9f5b9c9cf454f3845fc9af4881/certifi-2026.2.25-py3-none-any.whl", hash = "sha256:027692e4402ad994f1c42e52a4997a9763c646b73e4096e4d5d6db8af1d6f0fa", size = 153684, upload-time = "2026-02-25T02:54:15.766Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -267,20 +267,20 @@ wheels = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.2.0"
+version = "4.3.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "langchain-core" },
    { name = "ormsgpack" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/dc/e1/089c4c9e0a2fec7f883f82ae8e6a727138d50074cfeb6644bc2d13b1019b/langgraph_checkpoint-4.2.0.tar.gz", hash = "sha256:51a593b6bee684b0818e5d6e58e28ab340c6db7794575056ce7bd1b746a84ed7", size = 180239, upload-time = "2026-08-07T20:05:03.756Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/0f/69/31fdbdc65a85bbd6178afa193c772bb926620f47b4869638bc2bc80afaaa/langgraph_checkpoint-4.3.0.tar.gz", hash = "sha256:c75965d84cc2c1d549163e910a15bcb577758001b141619d05297c463280b018", size = 182652, upload-time = "2026-10-12T22:26:31.478Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/05/71/3b475f09bd57d3a5649792c66353312b4432afd843f301739dfcebd157f0/langgraph_checkpoint-4.2.0-py3-none-any.whl", hash = "sha256:0547fd228935a0b758865de3a3d6d7a2537c308895d0f9ab092ce9151b5da942", size = 56833, upload-time = "2026-08-07T20:05:02.655Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/0c/84747e340bf4f29291c84cdd5733fc8d0a822f3d33bb24e664a18afa4a7c/langgraph_checkpoint-4.3.0-py3-none-any.whl", hash = "sha256:bedfafe2f997ded60e4fa593e79f56f436a6e45586392dc382aa810d0c751c64", size = 58063, upload-time = "2026-10-12T22:26:30.429Z" },
 ]

 [[package]]
 name = "langgraph-checkpoint-postgres"
-version = "3.1.1"
+version = "3.1.3"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "langgraph-checkpoint" },
@@ -288,9 +288,9 @@ dependencies = [
    { name = "psycopg" },
    { name = "psycopg-pool" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/06/92/1e8959f8cd1b56e672fde3227f6fd642be85af6c5fd662d73921074aa39d/langgraph_checkpoint_postgres-3.1.1.tar.gz", hash = "sha256:d320e147ddad8c374cd546df0b52b532dd54d0541dd9fd23fc738cbd5de76f41", size = 150413, upload-time = "2026-07-30T19:15:39.014Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/78/bf/d0ab4d6e4d61952de2f77044d7407b7ce09e07d53e7bb448cf9df55c35e5/langgraph_checkpoint_postgres-3.1.3.tar.gz", hash = "sha256:a152a9c0c3d5931bc949b64e01e8c7da20be57a32aa754626446318e90a07650", size = 158122, upload-time = "2026-10-12T23:05:19.759Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/03/32/ba457698a48a0e18d786caa770033067049fbe36d6846f8e50f13b594b51/langgraph_checkpoint_postgres-3.1.1-py3-none-any.whl", hash = "sha256:6e353aecd8150de144fef8e51a49076f58b7d6830d4cf51392b7ad4d79832ba7", size = 50778, upload-time = "2026-07-30T19:15:37.405Z" },
+    { url = "https://files.pythonhosted.org/packages/41/42/659106ed829ee026144e32ddd589735f978d2ed09681020e5965cdfca04c/langgraph_checkpoint_postgres-3.1.3-py3-none-any.whl", hash = "sha256:050ae583223e24d97747f27b9e06e7345bf13c972f1fb6ed33bb3d1f9c11cee4", size = 52048, upload-time = "2026-10-12T23:05:18.854Z" },
 ]

 [[package]]
@@ -26,11 +26,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.5.20"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/f3/ce/ee2ecad540810a79593028e88299baeae54d346cc7a0d94b6199988b89b1/certifi-2026.5.20.tar.gz", hash = "sha256:69dea482ab64caa7b9f6aba1c6bf48bb6a5448d1c0f1b17ab42ad8c763a5344d", size = 135422, upload-time = "2026-05-20T11:46:50.073Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/59/8c/57e832b7af6d7c5abe66eb3fbe3a3a32f4d11ea23a1aa7131371035be991/certifi-2026.5.20-py3-none-any.whl", hash = "sha256:3c52e209ba0a4ad7aebe60436a4ab349c39e1e602e8c134221e546902ad25897", size = 134134, upload-time = "2026-05-20T11:46:48.578Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -1,10 +1,10 @@
 version = 1
-revision = 3
+revision = 5
 requires-python = ">=3.10"
 resolution-markers = [
    "python_full_version >= '3.15'",
    "python_full_version >= '3.12' and python_full_version < '3.15' and sys_platform == 'emscripten'",
-    "(python_full_version >= '3.11' and python_full_version < '3.15' and sys_platform != 'emscripten') or (python_full_version == '3.11.*' and sys_platform == 'emscripten')",
+    "python_full_version == '3.11.*' or (python_full_version >= '3.12' and python_full_version < '3.15' and sys_platform != 'emscripten')",
    "python_full_version < '3.11'",
 ]

@@ -24,7 +24,8 @@ source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "exceptiongroup", marker = "python_full_version < '3.11'" },
    { name = "idna" },
-    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11' and python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.16.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/61/cc/a381afa6efea9f496eff839d4a6a1aed3bfafc7b3ab4b0d1b243a12573dd/anyio-4.14.2.tar.gz", hash = "sha256:cfa139f3ed1a23ee8f88a145ddb5ac7605b8bbfd8592baacd7ce3d8bb4313c7f", size = 260176, upload-time = "2026-07-12T20:29:07.082Z" }
 wheels = [
@@ -168,11 +169,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.2.25"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/af/2d/7bf41579a8986e348fa033a31cdd0e4121114f6bce2457e8876010b092dd/certifi-2026.2.25.tar.gz", hash = "sha256:e887ab5cee78ea814d3472169153c2d12cd43b14bd03329a39a9c6e2e80bfba7", size = 155029, upload-time = "2026-02-25T02:54:17.342Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/9a/3c/c17fb3ca2d9c3acff52e30b309f538586f9f5b9c9cf454f3845fc9af4881/certifi-2026.2.25-py3-none-any.whl", hash = "sha256:027692e4402ad994f1c42e52a4997a9763c646b73e4096e4d5d6db8af1d6f0fa", size = 153684, upload-time = "2026-02-25T02:54:15.766Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -416,7 +417,7 @@ version = "50.0.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "cffi", marker = "platform_python_implementation != 'PyPy'" },
-    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+    { name = "typing-extensions", version = "4.16.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/de/41/6cbdcf9142d00fe82836fbb51e503e58088575cf7a0fe1dbff6695bf0840/cryptography-50.0.0.tar.gz", hash = "sha256:eeac2acb5a20ed25e0ad6d1df9891a520b78b404266b6d11778f25d5d691a6c9", size = 880201, upload-time = "2026-07-31T14:25:10.11Z" }
 wheels = [
@@ -469,11 +470,11 @@ wheels = [

 [[package]]
 name = "distlib"
-version = "0.4.0"
+version = "0.4.4"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/96/8e/709914eb2b5749865801041647dc7f4e6d00b549cfe88b65ca192995f07c/distlib-0.4.0.tar.gz", hash = "sha256:feec40075be03a04501a973d81f633735b4b69f98b05450592310c0f401a4e0d", size = 614605, upload-time = "2025-07-17T16:52:00.465Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/68/bc/d36966de7304047a1001a7559fbf266c464bb5d8cc85464536ef8c9bcc32/distlib-0.4.4.tar.gz", hash = "sha256:ca4175b4ac7f71223cc30ed7248c45d2b38c2439d8ee6e498ab67ed63cf3508d", size = 621304, upload-time = "2026-10-16T13:56:34.203Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/33/6b/e0547afaf41bf2c42e52430072fa5658766e3d65bd4b03a563d1b6336f57/distlib-0.4.0-py2.py3-none-any.whl", hash = "sha256:9659f7d87e46584a30b5780e43ac7a2143098441670ff0a49d5f9034c54a6c16", size = 469047, upload-time = "2025-07-17T16:51:58.613Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/2b/238f02a4927fa2e27aa995371d240a468b52e42eae44a678434aff736b38/distlib-0.4.4-py2.py3-none-any.whl", hash = "sha256:cf09d55b1e85a89576b036aeb71550bfd68320f3f21761ca329eadadad918638", size = 472099, upload-time = "2026-10-16T13:56:32.053Z" },
 ]

 [[package]]
@@ -496,7 +497,7 @@ name = "exceptiongroup"
 version = "1.3.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.16.0", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/50/79/66800aadf48771f6b62f7eb014e352e5d06856655206165d775e675a02c9/exceptiongroup-1.3.1.tar.gz", hash = "sha256:8b412432c6055b0b7d14c310000ae93352ed6754f70fa8f7c34141f91c4e3219", size = 30371, upload-time = "2025-11-21T23:01:54.787Z" }
 wheels = [
@@ -535,7 +536,7 @@ name = "grpcio"
 version = "1.80.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/b7/48/af6173dbca4454f4637a4678b67f52ca7e0c1ed7d5894d89d434fecede05/grpcio-1.80.0.tar.gz", hash = "sha256:29aca15edd0688c22ba01d7cc01cb000d72b2033f4a3c72a81a19b56fd143257", size = 12978905, upload-time = "2026-03-30T08:49:10.502Z" }
 wheels = [
@@ -811,7 +812,8 @@ dependencies = [
    { name = "httpx2-jsfetch", marker = "python_full_version >= '3.12' and sys_platform == 'emscripten'" },
    { name = "idna" },
    { name = "truststore", marker = "sys_platform != 'emscripten'" },
-    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11' and python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.16.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/7f/f8/579a8b51e42e38ee32647df9f08aa25643ae788e275cc625b199829c4671/httpx2-2.12.0.tar.gz", hash = "sha256:7631fe9887a8a2275f4a2540e053aa670fcc50742864a9ae7c66e609fdcf12cf", size = 100040, upload-time = "2026-08-18T13:22:09.086Z" }
 wheels = [
@@ -1000,7 +1002,7 @@ dependencies = [
    { name = "pydantic" },
    { name = "pyyaml" },
    { name = "tenacity" },
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
    { name = "uuid-utils" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/59/de/679a53472c25860837e32c0442c962fa86e95317a36460e2c9d5c91b17c2/langchain_core-1.4.0.tar.gz", hash = "sha256:1dc341eed802ed9c117c0df3923c991e5e9e226571e5725c194eeb5bd93d1a7f", size = 920260, upload-time = "2026-05-11T18:42:35.919Z" }
@@ -1013,7 +1015,7 @@ name = "langchain-protocol"
 version = "0.0.17"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/44/b3/4e2429876c7a35585618caa2b9f9089f7162a6b50562b614ad82ac11c17e/langchain_protocol-0.0.17.tar.gz", hash = "sha256:e7cbe58c205df4b4fd87dc6d5bb23f10e13b236d0e2e1b0b9d05bc2b648f3eea", size = 6026, upload-time = "2026-06-12T18:39:51.923Z" }
 wheels = [
@@ -1337,84 +1339,114 @@ wheels = [

 [[package]]
 name = "opentelemetry-api"
-version = "1.41.0"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "importlib-metadata" },
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/47/8e/3778a7e87801d994869a9396b9fc2a289e5f9be91ff54a27d41eace494b0/opentelemetry_api-1.41.0.tar.gz", hash = "sha256:9421d911326ec12dee8bc933f7839090cad7a3f13fcfb0f9e82f8174dc003c09", size = 71416, upload-time = "2026-04-09T14:38:34.544Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/2e/02/6e0ae9cc61bd3169d401077b507b3ebc344745171e1051ab430be012dcd9/opentelemetry_api-1.45.1.tar.gz", hash = "sha256:aa38ed19bcc084ba42782a73255b3582283eced7ad6dddbd6695189e69adfb75", size = 72804, upload-time = "2026-10-06T17:32:58.133Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/58/ee/99ab786653b3bda9c37ade7e24a7b607a1b1f696063172768417539d876d/opentelemetry_api-1.41.0-py3-none-any.whl", hash = "sha256:0e77c806e6a89c9e4f8d372034622f3e1418a11bdbe1c80a50b3d3397ad0fa4f", size = 69007, upload-time = "2026-04-09T14:38:11.833Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/41/f7dcf80b81ee8e71c1a2b59f14208bc723edbd89ed027a73b175abf6348e/opentelemetry_api-1.45.1-py3-none-any.whl", hash = "sha256:b31553efa588ae44bc306f863c785c5333a9ecc091248c6ee68b4b6c87fdedfb", size = 60256, upload-time = "2026-10-06T17:32:33.506Z" },
+]
+
+[[package]]
+name = "opentelemetry-exporter-http-transport"
+version = "0.66b1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "opentelemetry-api" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/62/0c/e3ebdb4b507f66afcc905e6885a4946969bd75b45988492643356fbbdc63/opentelemetry_exporter_http_transport-0.66b1.tar.gz", hash = "sha256:443080203bf52586ce0b2ad901e8951c61833eab1aa539ae6f1f16fe9e8e7952", size = 11693, upload-time = "2026-10-06T17:32:59.65Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/04/69/6af86ff66492b481c6a4c05dcfd68beb47ed8ba046440a26a2aac76b95c7/opentelemetry_exporter_http_transport-0.66b1-py3-none-any.whl", hash = "sha256:2f95404bdee7f9d2d529c7de56c7bd86d014d774d8fbf137810e0167f8a492bf", size = 12155, upload-time = "2026-10-06T17:32:35.454Z" },
+]
+
+[package.optional-dependencies]
+requests = [
+    { name = "requests" },
+]
+
+[[package]]
+name = "opentelemetry-exporter-otlp-common"
+version = "0.66b1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "opentelemetry-sdk" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/cb/19/41de712173f43057e4532d42ece7d0c6d4210d353e5752433cb14987643f/opentelemetry_exporter_otlp_common-0.66b1.tar.gz", hash = "sha256:6b1403487a2185ac1feb45fd5546fdf8630ce71c36bcefaadf51e2130e9e23f9", size = 14325, upload-time = "2026-10-06T17:33:01.725Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fc/39/8c23d67665c762aa51840fa06f86e902e8f6f1693bc8d7e3d98cd6e2f753/opentelemetry_exporter_otlp_common-0.66b1-py3-none-any.whl", hash = "sha256:00ff8592c3a7cb729ff3fdc7ffa12372c243bdf2163e80c180994d0c7bd83ee9", size = 12385, upload-time = "2026-10-06T17:32:38.177Z" },
 ]

 [[package]]
 name = "opentelemetry-exporter-otlp-proto-common"
-version = "1.41.0"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "opentelemetry-proto" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/8c/28/e8eca94966fe9a1465f6094dc5ddc5398473682180279c94020bc23b4906/opentelemetry_exporter_otlp_proto_common-1.41.0.tar.gz", hash = "sha256:966bbce537e9edb166154779a7c4f8ab6b8654a03a28024aeaf1a3eacb07d6ee", size = 20411, upload-time = "2026-04-09T14:38:36.572Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/c1/8e/65e85e5137991a3c493b11682151d198638a5bc1dd4b4c5f67e013c57d7c/opentelemetry_exporter_otlp_proto_common-1.45.1.tar.gz", hash = "sha256:2e4adcc3a67bcf57804fc49514f0ef64974ca7590aa3491da389852b4a0628f6", size = 18873, upload-time = "2026-10-06T17:33:04.471Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/26/c4/78b9bf2d9c1d5e494f44932988d9d91c51a66b9a7b48adf99b62f7c65318/opentelemetry_exporter_otlp_proto_common-1.41.0-py3-none-any.whl", hash = "sha256:7a99177bf61f85f4f9ed2072f54d676364719c066f6d11f515acc6c745c7acf0", size = 18366, upload-time = "2026-04-09T14:38:15.135Z" },
+    { url = "https://files.pythonhosted.org/packages/84/aa/92f225d353904e7f70b8b3e3c1b02db0cf56f744c2e83c581dc372e78873/opentelemetry_exporter_otlp_proto_common-1.45.1-py3-none-any.whl", hash = "sha256:2f446183ae7047b036226f1d846c41a834b0e8755ad13b51a51dd38952eb466c", size = 15393, upload-time = "2026-10-06T17:32:41.911Z" },
 ]

 [[package]]
 name = "opentelemetry-exporter-otlp-proto-http"
-version = "1.41.0"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "googleapis-common-protos" },
    { name = "opentelemetry-api" },
+    { name = "opentelemetry-exporter-http-transport", extra = ["requests"] },
+    { name = "opentelemetry-exporter-otlp-common" },
    { name = "opentelemetry-exporter-otlp-proto-common" },
    { name = "opentelemetry-proto" },
    { name = "opentelemetry-sdk" },
    { name = "requests" },
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/19/63/d9f43cd75f3fabb7e01148c89cfa9491fc18f6580a6764c554ff7c953c46/opentelemetry_exporter_otlp_proto_http-1.41.0.tar.gz", hash = "sha256:dcd6e0686f56277db4eecbadd5262124e8f2cc739cadbc3fae3d08a12c976cf5", size = 24139, upload-time = "2026-04-09T14:38:38.128Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/1b/17/26487707ea4caa97b17e6e4b5fa72133a53512ffa2f5cf7a49ef284b29cb/opentelemetry_exporter_otlp_proto_http-1.45.1.tar.gz", hash = "sha256:45c218405ce3fd879596924b1874bf9a8f6880206d61065c5a912c8e5c297fb7", size = 28839, upload-time = "2026-10-06T17:33:05.713Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/64/b5/a214cd907eedc17699d1c2d602288ae17cb775526df04db3a3b3585329d2/opentelemetry_exporter_otlp_proto_http-1.41.0-py3-none-any.whl", hash = "sha256:a9c4ee69cce9c3f4d7ee736ad1b44e3c9654002c0816900abbafd9f3cf289751", size = 22673, upload-time = "2026-04-09T14:38:18.349Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/1f/517eaa0187ba106a9da97160ce2add3a371812681dc440930b267f714e42/opentelemetry_exporter_otlp_proto_http-1.45.1-py3-none-any.whl", hash = "sha256:24a97cf3753c7fb52fad44a696e452ff371686339e2acf3309e2eda3d0230700", size = 22180, upload-time = "2026-10-06T17:32:43.946Z" },
 ]

 [[package]]
 name = "opentelemetry-proto"
-version = "1.41.0"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "protobuf" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/e0/d9/08e3dc6156878713e8c811682bc76151f5fe1a3cb7f3abda3966fd56e71e/opentelemetry_proto-1.41.0.tar.gz", hash = "sha256:95d2e576f9fb1800473a3e4cfcca054295d06bdb869fda4dc9f4f779dc68f7b6", size = 45669, upload-time = "2026-04-09T14:38:45.978Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/4b/7f/15f014fb195da6c2dbb6c71399b8e76824878718e94de6454038488eed28/opentelemetry_proto-1.45.1.tar.gz", hash = "sha256:79e0fb95e4616691a469439238aa9224d75779b3e108e895d1aa125ab29ca77c", size = 46488, upload-time = "2026-10-06T17:33:11.49Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/49/8c/65ef7a9383a363864772022e822b5d5c6988e6f9dabeebb9278f5b86ebc3/opentelemetry_proto-1.41.0-py3-none-any.whl", hash = "sha256:b970ab537309f9eed296be482c3e7cca05d8aca8165346e929f658dbe153b247", size = 72074, upload-time = "2026-04-09T14:38:29.38Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/9a/42ec8180a769516ae757e893b69736826efceac7332553915b4528a91c6d/opentelemetry_proto-1.45.1-py3-none-any.whl", hash = "sha256:f38e2a8413053c180cd3d2637fbb279673ec2f6a6e09c995aafa2f452c52b46e", size = 72488, upload-time = "2026-10-06T17:32:53.057Z" },
 ]

 [[package]]
 name = "opentelemetry-sdk"
-version = "1.41.0"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "opentelemetry-api" },
    { name = "opentelemetry-semantic-conventions" },
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/f8/0e/a586df1186f9f56b5a0879d52653effc40357b8e88fc50fe300038c3c08b/opentelemetry_sdk-1.41.0.tar.gz", hash = "sha256:7bddf3961131b318fc2d158947971a8e37e38b1cd23470cfb72b624e7cc108bd", size = 230181, upload-time = "2026-04-09T14:38:47.225Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a1/79/7392e21a1c8f0c61d90b223e31c7e48cb9d452e91a6b820ad24cca5f23c4/opentelemetry_sdk-1.45.1.tar.gz", hash = "sha256:63d24a6ca645019a631e6a51999c73e93adcac1196ca640b8ae78a7cc4762bf3", size = 218324, upload-time = "2026-10-06T17:33:13.26Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/2c/13/a7825118208cb32e6a4edcd0a99f925cbef81e77b3b0aedfd9125583c543/opentelemetry_sdk-1.41.0-py3-none-any.whl", hash = "sha256:a596f5687964a3e0d7f8edfdcf5b79cbca9c93c7025ebf5fb00f398a9443b0bd", size = 180214, upload-time = "2026-04-09T14:38:30.657Z" },
+    { url = "https://files.pythonhosted.org/packages/95/3c/87c42b4bd6dd297536f04cd9383d212ac557ecd49f2cbdcd46da1c9ef5c8/opentelemetry_sdk-1.45.1-py3-none-any.whl", hash = "sha256:c604c11dc429810812348989115fa44bd558772a3d7442afc43d024f2c250ca4", size = 140063, upload-time = "2026-10-06T17:32:55.04Z" },
 ]

 [[package]]
 name = "opentelemetry-semantic-conventions"
-version = "0.62b0"
+version = "0.66b1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "opentelemetry-api" },
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/a3/b0/c14f723e86c049b7bf8ff431160d982519b97a7be2857ed2247377397a24/opentelemetry_semantic_conventions-0.62b0.tar.gz", hash = "sha256:cbfb3c8fc259575cf68a6e1b94083cc35adc4a6b06e8cf431efa0d62606c0097", size = 145753, upload-time = "2026-04-09T14:38:48.274Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/46/e4/dbbfb2a010c4db2224a5114638acede6fe563d33cc20fb1752cebcbe6298/opentelemetry_semantic_conventions-0.66b1.tar.gz", hash = "sha256:497ca63bf383723411e8eaf60c8779e9877633c936bb641080adab59d0eb6ec8", size = 150250, upload-time = "2026-10-06T17:33:14.073Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/58/6c/5e86fa1759a525ef91c2d8b79d668574760ff3f900d114297765eb8786cb/opentelemetry_semantic_conventions-0.62b0-py3-none-any.whl", hash = "sha256:0ddac1ce59eaf1a827d9987ab60d9315fb27aea23304144242d1fcad9e16b489", size = 231619, upload-time = "2026-04-09T14:38:32.394Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/14/67f8aa798857f8cf686f515bf93d9bb877ce952ddc8efae0fa25b45ce0d6/opentelemetry_semantic_conventions-0.66b1-py3-none-any.whl", hash = "sha256:d4cddeb4315490b35213f55e2bdc9ac54bb1e4d318927475bed62b35545e581b", size = 206279, upload-time = "2026-10-06T17:32:56.103Z" },
 ]

 [[package]]
@@ -1556,11 +1588,11 @@ wheels = [

 [[package]]
 name = "packaging"
-version = "26.1"
+version = "26.3"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/df/de/0d2b39fb4af88a0258f3bac87dfcbb48e73fbdea4a2ed0e2213f9a4c2f9a/packaging-26.1.tar.gz", hash = "sha256:f042152b681c4bfac5cae2742a55e103d27ab2ec0f3d88037136b6bfe7c9c5de", size = 215519, upload-time = "2026-04-14T21:12:49.362Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/7d/fa/3944b40b07da9ce895c0e6303a5ab7d53da063554f534556b134a54d6093/packaging-26.3.tar.gz", hash = "sha256:94edc256424af38762eb31306eed28beb9f0efc50a8837492c9d6fd6004aed79", size = 313412, upload-time = "2026-08-04T18:15:28.737Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/7a/c2/920ef838e2f0028c8262f16101ec09ebd5969864e5a64c4c05fad0617c56/packaging-26.1-py3-none-any.whl", hash = "sha256:5d9c0669c6285e491e0ced2eee587eaf67b670d94a19e94e3984a481aba6802f", size = 95831, upload-time = "2026-04-14T21:12:47.56Z" },
+    { url = "https://files.pythonhosted.org/packages/63/34/ba1c580383c9eada3711951fef0795c80b829a078d72188184bcab9dd527/packaging-26.3-py3-none-any.whl", hash = "sha256:d7193f7c8e4e93f444fde0262bf90af30e16fa0ad0ad44cb553c87339b23cd1c", size = 129956, upload-time = "2026-08-04T18:15:27.159Z" },
 ]

 [[package]]
@@ -1642,7 +1674,7 @@ source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "annotated-types" },
    { name = "pydantic-core" },
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
    { name = "typing-inspection" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/09/e5/06d23afac9973109d1e3c8ad38e1547a12e860610e327c05ee686827dc37/pydantic-2.13.2.tar.gz", hash = "sha256:b418196607e61081c3226dcd4f0672f2a194828abb9109e9cfb84026564df2d1", size = 843836, upload-time = "2026-04-17T09:31:59.636Z" }
@@ -1655,7 +1687,7 @@ name = "pydantic-core"
 version = "2.46.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/43/bb/4742f05b739b2478459bb16fa8470549518c802e06ddcf3f106c5081315e/pydantic_core-2.46.2.tar.gz", hash = "sha256:37bb079f9ee3f1a519392b73fda2a96379b31f2013c6b467fe693e7f2987f596", size = 471269, upload-time = "2026-04-17T09:10:07.017Z" }
 wheels = [
@@ -1818,7 +1850,8 @@ source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "backports-asyncio-runner", marker = "python_full_version < '3.11'" },
    { name = "pytest" },
-    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11' and python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.16.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/43/7c/d36d04db312ecf4298932ef77e6e4a9e8ad017906e24e34f0b0c361a2473/pytest_asyncio-1.4.0.tar.gz", hash = "sha256:c6c0d2259945122819f171a32ecea2c349ead889ee28176caaf492143424be42", size = 58514, upload-time = "2026-05-26T09:56:04.083Z" }
 wheels = [
@@ -2035,11 +2068,11 @@ wheels = [

 [[package]]
 name = "setuptools"
-version = "83.0.0"
+version = "84.1.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/34/26/f5d29e25ffdb535afef2d35cdb55b325298f96debd670da4c325e08d70f4/setuptools-83.0.0.tar.gz", hash = "sha256:025bccbbf0fa05b6192bc64ae1e7b16e001fd6d6d4d5de03c97b1c1ade523bef", size = 1154254, upload-time = "2026-07-04T15:31:22.699Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/73/eb/7e5f628ee2eb168fecbf8026de7856b2543d6d609cc85f086b9a91ac93d6/setuptools-84.1.0.tar.gz", hash = "sha256:23657205b63d4d31395ab917815247f694aafad28cd21316bf5fd45a51048ea0", size = 1171783, upload-time = "2026-10-19T20:11:43.489Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/5d/40/e1e72872c6354b306daef1703549e8e83b4d43cfea356311bf722a043752/setuptools-83.0.0-py3-none-any.whl", hash = "sha256:29b23c360f22f414dc7336bb39178cc7bcbf6021ed2733cde173f09dba19abb3", size = 1008090, upload-time = "2026-07-04T15:31:20.885Z" },
+    { url = "https://files.pythonhosted.org/packages/18/02/b4bbb90f8f29e334898ff45c6d9b1fffaa6f378017052b3023a06262b5f7/setuptools-84.1.0-py3-none-any.whl", hash = "sha256:5231689c6ac8156cd155cfb98c3fd6fb252c3e7e9bd3d78b64d3c55f132624c9", size = 819380, upload-time = "2026-10-19T20:11:41.684Z" },
 ]

 [[package]]
@@ -2079,7 +2112,7 @@ version = "1.3.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "anyio" },
-    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.13'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/eb/e3/7c1dc7381d9f8ab7d854328ebfa884e62cb3f3d8549ddfd37c7814f42afa/starlette-1.3.1.tar.gz", hash = "sha256:05d0213193f2fbaae60e2ecb593b4add4262ad4e46536b54abe36f11a71724e0", size = 2703240, upload-time = "2026-06-12T09:23:11.602Z" }
 wheels = [
@@ -2223,17 +2256,34 @@ wheels = [
 name = "typing-extensions"
 version = "4.15.0"
 source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version >= '3.15'",
+    "python_full_version >= '3.12' and python_full_version < '3.15' and sys_platform == 'emscripten'",
+    "python_full_version == '3.11.*' or (python_full_version >= '3.12' and python_full_version < '3.15' and sys_platform != 'emscripten')",
+]
 sdist = { url = "https://files.pythonhosted.org/packages/72/94/1a15dd82efb362ac84269196e94cf00f187f7ed21c242792a923cdb1c61f/typing_extensions-4.15.0.tar.gz", hash = "sha256:0cea48d173cc12fa28ecabc3b837ea3cf6f38c6d1136f85cbaaf598984861466", size = 109391, upload-time = "2025-08-25T13:49:26.313Z" }
 wheels = [
    { url = "https://files.pythonhosted.org/packages/18/67/36e9267722cc04a6b9f15c7f3441c2363321a3ea07da7ae0c0707beb2a9c/typing_extensions-4.15.0-py3-none-any.whl", hash = "sha256:f0fa19c6845758ab08074a0cfa8b7aecb71c999ca73d62883bc25cc018c4e548", size = 44614, upload-time = "2025-08-25T13:49:24.86Z" },
 ]

+[[package]]
+name = "typing-extensions"
+version = "4.16.0"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version < '3.11'",
+]
+sdist = { url = "https://files.pythonhosted.org/packages/f6/cc/6253133b5bb138fc3306cebfbda2c520f545d36b5be2c7255cc528bb45d6/typing_extensions-4.16.0.tar.gz", hash = "sha256:dc983d19a509c94dba722ee6abd33940f7c05a89e243c47e907eb4db6f1a43e5", size = 113555, upload-time = "2026-07-02T08:40:05.92Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/49/d3/b8441a820a491ddfc024b0b0cf0393375b75ea13866d9c66727e54c2fc80/typing_extensions-4.16.0-py3-none-any.whl", hash = "sha256:481caa481374e813c1b176ada14e97f1f67a4539ce9cfeb3f350d78d6370c2e8", size = 45571, upload-time = "2026-07-02T08:40:04.659Z" },
+]
+
 [[package]]
 name = "typing-inspection"
 version = "0.4.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions" },
+    { name = "typing-extensions", version = "4.15.0", source = { registry = "https://pypi.org/simple" } },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/55/e3/70399cb7dd41c10ac53367ae42139cf4b1ca5f36bb3dc6c9d33acdb43655/typing_inspection-0.4.2.tar.gz", hash = "sha256:ba561c48a67c5958007083d386c3295464928b01faa735ab8547c5692e87f464", size = 75949, upload-time = "2025-10-01T02:14:41.687Z" }
 wheels = [
@@ -2292,28 +2342,28 @@ wheels = [

 [[package]]
 name = "uv"
-version = "0.11.15"
+version = "0.13.4"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/da/34/609d5d01ba21dc8f0974610ca7802fbb2c946a0c38665cfe5c5aeddbefb5/uv-0.11.15.tar.gz", hash = "sha256:755f959ec6a2fd8ccb6ee76ad90ab759d2eb1f4797444078645dd1ee4bca92d6", size = 4159545, upload-time = "2026-05-18T19:57:48.133Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/79/55/609455d21e87430c11fe734abfced50a3bdf177b2f49d41fbb19c199cbc3/uv-0.13.4.tar.gz", hash = "sha256:b0bff5b9a5061f7a35b1d0c1a94c19af134222c48eb7aadc42f85a038be1b248", size = 7482486, upload-time = "2026-10-20T15:12:06.801Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/a6/7c/dcc230c5911884d8848145dabcac8fb95a5ed6f9fe1c57fae8242618f28a/uv-0.11.15-py3-none-linux_armv6l.whl", hash = "sha256:83b04ab49514a0a761ffedb36a748ee81f87746671e72088e5f32c9585e5f1a9", size = 23110183, upload-time = "2026-05-18T19:57:23.051Z" },
-    { url = "https://files.pythonhosted.org/packages/f4/f3/efd4e044b60eb9c3c12ee386be098d56c335538ccec7caa49349cfba9344/uv-0.11.15-py3-none-macosx_10_12_x86_64.whl", hash = "sha256:b6cae61f737be075b90be9e3f07d961072aed7019f4c9b8ed5c5d41c4d6cade3", size = 22637941, upload-time = "2026-05-18T19:57:26.752Z" },
-    { url = "https://files.pythonhosted.org/packages/a6/b8/48627f895a1569e576822e0a8416aa4797eb4a4551de21a4ad97b9b5819d/uv-0.11.15-py3-none-macosx_11_0_arm64.whl", hash = "sha256:9accae33619a9166e5c48531deb455d672cfb89f9357a00975e669c76b0bd49f", size = 21258803, upload-time = "2026-05-18T19:57:05.473Z" },
-    { url = "https://files.pythonhosted.org/packages/af/50/4bc8a148274feabee2d9c9f1fa15009e10c0228dfe57981ee3ea2ef1d481/uv-0.11.15-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.musllinux_1_1_aarch64.whl", hash = "sha256:c0cf52cd6d50bb9e05e2d968f45f80761107e4cbc8d4a26d9758f9d8274aaec1", size = 23066178, upload-time = "2026-05-18T19:57:33.058Z" },
-    { url = "https://files.pythonhosted.org/packages/a9/56/139fc3bec9a8b0a25bfe2196123adb9f16124da437bf4fbcf0d21cfcafb2/uv-0.11.15-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.musllinux_1_1_armv7l.whl", hash = "sha256:49dc6ed70bff00937384f96cdc4b1a4742d18e5504ec2c4a1214dba2dee5687a", size = 22705332, upload-time = "2026-05-18T19:57:36.714Z" },
-    { url = "https://files.pythonhosted.org/packages/ca/b0/b18b3dd204f8c213236a1ebd148e009861637129a8cce34df0e9aa22ed40/uv-0.11.15-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:adb9a89352539fdd8f7cd5f9966cf9f94fc5b98e0ccdf5003a04123dc6423bec", size = 22707534, upload-time = "2026-05-18T19:58:04.117Z" },
-    { url = "https://files.pythonhosted.org/packages/76/36/3ca09f95572df99d361b49c96b1297149e96e120d8d1ecf074095a4b6da4/uv-0.11.15-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:40ff67e3f8e8a7533781a2e892a534975a93acb83ea35460e64e7b2bf2111774", size = 24096607, upload-time = "2026-05-18T19:58:11.625Z" },
-    { url = "https://files.pythonhosted.org/packages/64/be/3bdee21a296bbf5336a526e3613d0e7d4538dacc39c62d7fcba55d15f6b0/uv-0.11.15-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:c6463a299ed7e6b5a800ed6f108af8e1588352629424133ddef7572b0e1e1118", size = 25082562, upload-time = "2026-05-18T19:57:40.69Z" },
-    { url = "https://files.pythonhosted.org/packages/cd/73/f371f3689ffe741066468d001d85f739fc4b5574de83b639ef19b5e8a7f4/uv-0.11.15-py3-none-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:68c1e62d4b78578b90b833553286b65d6a7e327537716441068583ba652ec4f5", size = 24253391, upload-time = "2026-05-18T19:57:18.47Z" },
-    { url = "https://files.pythonhosted.org/packages/d3/16/fe392d618af6b00c064b3e718d585dcf791546a77c5123a5bec07ce53a0a/uv-0.11.15-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:98edf1bdaf82447014852051d93e3ee95012509c567bf057fd117e6bdbd9a807", size = 24415871, upload-time = "2026-05-18T19:58:19.651Z" },
-    { url = "https://files.pythonhosted.org/packages/6e/24/2e92a052fb6334fcd746d1c7cb57847c204b118c84f5da53c0f9e129f7b7/uv-0.11.15-py3-none-manylinux_2_28_aarch64.whl", hash = "sha256:be8f76d25bcf4c92bb384240ac1bf9aa7f51063d0bdeca4c9cf0ec3ed8b145e0", size = 23159007, upload-time = "2026-05-18T19:57:10.653Z" },
-    { url = "https://files.pythonhosted.org/packages/3d/2e/6923d0658d164bb2c435ed1868aa2d49b3074594679917a001ff92dc95bb/uv-0.11.15-py3-none-manylinux_2_31_riscv64.musllinux_1_1_riscv64.whl", hash = "sha256:f9f4fbbf4fe485522054f3c7496c6e8e932d6436e4200ff3daf718db0b7c7bd5", size = 23769385, upload-time = "2026-05-18T19:58:15.856Z" },
-    { url = "https://files.pythonhosted.org/packages/a4/99/7e34cd949e57360814e8064cc9fb7104df445d0f6a663504e5f7473480aa/uv-0.11.15-py3-none-manylinux_2_31_riscv64.whl", hash = "sha256:0ed920e896b2fd13a35031707e307e42fbb2681458b967440a17272d86d49137", size = 23860973, upload-time = "2026-05-18T19:57:55.575Z" },
-    { url = "https://files.pythonhosted.org/packages/28/98/8fe1f5f9d816e94569a0298dd8e0936801097625fa1952162951f0d628b6/uv-0.11.15-py3-none-musllinux_1_1_i686.whl", hash = "sha256:41d907611f3e6a13262807fd7f0a17849f76285ca80f536f6b3943732bdc6656", size = 23431392, upload-time = "2026-05-18T19:57:59.814Z" },
-    { url = "https://files.pythonhosted.org/packages/cc/6b/76a1ce2fa860026913a5941700cdc7d715fce9c3277a3fa3489cf2523ca0/uv-0.11.15-py3-none-musllinux_1_1_x86_64.whl", hash = "sha256:e3b68f8bf1a4568710f77e5bda9182ce7682811d89a8e7468c22460e032b234d", size = 24519478, upload-time = "2026-05-18T19:57:51.165Z" },
-    { url = "https://files.pythonhosted.org/packages/43/60/1d58e8a05718cb50494763115710b73846cacb651fd735d285233fd72c59/uv-0.11.15-py3-none-win32.whl", hash = "sha256:8e2da3076761086a5b76869c3f38ef0509c836046ef41ddd19485dfd7271dca9", size = 22020178, upload-time = "2026-05-18T19:58:07.64Z" },
-    { url = "https://files.pythonhosted.org/packages/55/53/40fcefcb348af660488597ed3c01363df7344e60611f8883750dc596f5c6/uv-0.11.15-py3-none-win_amd64.whl", hash = "sha256:cc3915ab291a1ecaf31de05f5d8bd70d09c66fe9911a53f70d9efa62ff0dbd8a", size = 24668779, upload-time = "2026-05-18T19:57:44.894Z" },
-    { url = "https://files.pythonhosted.org/packages/e5/7d/fa3a9960c95af9bbe2a629048760d0b9b4fead8ccd4f2235af747ec7cdf0/uv-0.11.15-py3-none-win_arm64.whl", hash = "sha256:4f39426a13dee24897aed60c4b98058c66f18bd983885ac5f4a54a04b24fbddf", size = 23198178, upload-time = "2026-05-18T19:57:14.68Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/bd/4349da5356b17a1c0e36c8552c809925f86f6292de9b1f599f01e8e8093d/uv-0.13.4-py3-none-linux_armv6l.whl", hash = "sha256:205a4f93e568fb24f45a2f3834f1c0a8f1adb5f396bf473c71030ed707dce048", size = 18272390, upload-time = "2026-10-20T15:11:02.468Z" },
+    { url = "https://files.pythonhosted.org/packages/21/f2/0cb77a1ae43959772f8f7873eb1f4061907fbb1694ce55c79b231f639b1d/uv-0.13.4-py3-none-macosx_10_12_x86_64.whl", hash = "sha256:b4bbacea78c9db136cd2ec3147cbfd9ed910b4db5e1877109bd74d0e73bb2eaf", size = 17384113, upload-time = "2026-10-20T15:11:07.172Z" },
+    { url = "https://files.pythonhosted.org/packages/26/44/5a8733dd1b2e8d6f9cffd1ff6428a3e11dcc7e146a7657e32cfb8d9fb4a4/uv-0.13.4-py3-none-macosx_11_0_arm64.whl", hash = "sha256:95f4eb2b991b53b50ec2b18ab612fca8ba1e7a75a47af0331fb8bcd21da3eb03", size = 14433723, upload-time = "2026-10-20T15:11:10.185Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/39/be14f051bf34a1822d634819890a2d0cba3b69101e633fb7b9f54d7b4020/uv-0.13.4-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.musllinux_1_1_aarch64.whl", hash = "sha256:e90ab42825587821096cb0d071e387468da38cfa22818c91f9c0b3c0453048e8", size = 17816443, upload-time = "2026-10-20T15:11:13.674Z" },
+    { url = "https://files.pythonhosted.org/packages/97/a0/5df0f8bc6367dd37b98432a7c462e2918554f6db2a71211574852b8f0727/uv-0.13.4-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.musllinux_1_1_armv7l.whl", hash = "sha256:3a9dcb6762c3ba249d8c084315ba03839082af3ece3536cc4164ec735c351d68", size = 17828025, upload-time = "2026-10-20T15:11:16.806Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/9c/2fed5357700b7c304b94214979e4efe680c3bc6e337a97eedaa372fe0947/uv-0.13.4-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:8ae6ff2b07892379dc767c9064c2b8bbfac88e767edac7ce28e1b0f15d86abb4", size = 17843737, upload-time = "2026-10-20T15:11:20.043Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/2e/ee026284f096236dc3e1abed0316f9edab65c0ea176ea0072c064e62478d/uv-0.13.4-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:25cc14ff4fd8335d3be2cc493a621a003fdafcb0fb1c953c905daabe4def5ab8", size = 18576350, upload-time = "2026-10-20T15:11:24.035Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/f1/67fbdb649c8c5d0cb38ea74829080479b77c8631cda5d77fc1b28f027ca7/uv-0.13.4-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:77062b02856c74697f48328d8a6583ab07e6da0a993e86c2db46f3d1da27c637", size = 19831234, upload-time = "2026-10-20T15:11:27.627Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/00/7d49e3a1a3e98e5244fa38e0c964dc8de568bf3bb101257a63dc2602efed/uv-0.13.4-py3-none-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:fb0e2028d9a8929884344afb7d91d0f65d9783f640d28aede89fca3602bedde9", size = 19702886, upload-time = "2026-10-20T15:11:31.128Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/a1/de02c0913726728eca56ee258875cbba9c7b0572c63f494fd9f55c7852e3/uv-0.13.4-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:05e4332791b4d55b6dde8695b7fa6079b26991519f44760887c5dfbd5ca4fe53", size = 17181168, upload-time = "2026-10-20T15:11:34.583Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/cd/f4a3c7377674366798c56aabd0f9a1f05687c3939f5ba857a3ea6093ff84/uv-0.13.4-py3-none-manylinux_2_28_aarch64.whl", hash = "sha256:7b3cdf868e5415631fe797b648c0d6945d38d388f23367bdd0b6332e80a1fcb0", size = 16502020, upload-time = "2026-10-20T15:11:37.88Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/f4/ce79657c5a0f51b602db9a663529baafff03b9fe8cc41e09ad3f7246b88a/uv-0.13.4-py3-none-manylinux_2_31_riscv64.musllinux_1_1_riscv64.whl", hash = "sha256:4e6b08578c920b86910897f36bb6da7229c8e7e0187976570e2a538cb28b777e", size = 18000980, upload-time = "2026-10-20T15:11:41.59Z" },
+    { url = "https://files.pythonhosted.org/packages/21/e6/7d77022d5661f8e01578bc5423bf02af408220e7e7ac5ee15eb94742c7cb/uv-0.13.4-py3-none-manylinux_2_31_riscv64.whl", hash = "sha256:d86a5dec8ab30c70519c71ca5085c8aaf28664836cbc600951cedaa3fc678cbf", size = 18179083, upload-time = "2026-10-20T15:11:45.191Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/40/1bf0324800d38009774ba087a6fbffddc76a3e5bf42c081b9298b1236fb5/uv-0.13.4-py3-none-musllinux_1_1_i686.whl", hash = "sha256:5d89d1028c2e22977d9215cbb79cfe12559cb0c8a507a12e2f3a7e47930cf1bf", size = 18010389, upload-time = "2026-10-20T15:11:48.963Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/ec/f352ca2cc3b4b6191bb4ec8cdf4bc190eed0ba8bda4c33b7fd66c5d33cec/uv-0.13.4-py3-none-musllinux_1_1_x86_64.whl", hash = "sha256:52b5c753a7c2d5f8323a70bb4c93118988356331d0758afa6446c81f974e7196", size = 19004998, upload-time = "2026-10-20T15:11:52.628Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/4f/fc05ff4292edfe785b47874ec86d3d83ca2bf735cff8a9d2170ffb66ad83/uv-0.13.4-py3-none-win32.whl", hash = "sha256:d333f01852b817ba05612de31474739da1af7e5463d119b30a9c4331eba1f408", size = 16190502, upload-time = "2026-10-20T15:11:56.657Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/b9/b526eab6c72eacc705a58378f6850fde046b28c86fed01bd42da61367c6b/uv-0.13.4-py3-none-win_amd64.whl", hash = "sha256:b2ff80397273fff0e02c20eacae6882834c811827d6fc08f658db156fe754386", size = 15331833, upload-time = "2026-10-20T15:11:59.733Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/f7/4d5fa89d3d10ff6a84e547ef6e02c45b2f2d50e66682da778acf60db63f1/uv-0.13.4-py3-none-win_arm64.whl", hash = "sha256:deffa5a4afdb48f3e44fb26e80377d256b330ddcad0373daff4b54f6162b43a4", size = 15690151, upload-time = "2026-10-20T15:12:02.74Z" },
 ]

 [[package]]
@@ -2375,18 +2425,19 @@ wheels = [

 [[package]]
 name = "virtualenv"
-version = "21.7.13"
+version = "21.14.13"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "distlib" },
    { name = "filelock" },
+    { name = "packaging" },
    { name = "platformdirs" },
    { name = "python-discovery" },
-    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+    { name = "typing-extensions", version = "4.16.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/13/50/c9b84eb106d0db420b9878dac0a386c5791726f127ce20b06897f6e3a1e9/virtualenv-21.7.13.tar.gz", hash = "sha256:0355558b6f33619aab31347e43643b0ebc97f61ea3acf617b2b69e1f8a843d11", size = 5360306, upload-time = "2026-09-18T04:35:49.354Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/70/aa/bff50daee0488e16d611258a9b6c5a1d0c9e128f8ca424e9f4e1c0444ef2/virtualenv-21.14.13.tar.gz", hash = "sha256:de5307ba4b9a0001d9c8ea02832d38f63ca432880f416c1bface61c4323fd905", size = 5570301, upload-time = "2026-10-22T20:10:16.952Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/9a/ce/e74453531b0c49a58d0e8a4f9bab4495705859fee4a4f7de27d58f4a791a/virtualenv-21.7.13-py3-none-any.whl", hash = "sha256:1bea5af7463f59c4719db48fe739579a2a4f569c96f26c086edda85c96da9f59", size = 5328680, upload-time = "2026-09-18T04:35:47.194Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/a9/c10d8a44a29fb94a18b43fdf11f38caca53ae7c9406f11a72b1cb8f624f1/virtualenv-21.14.13-py3-none-any.whl", hash = "sha256:5fbb04138a4861580b78ba9110d529a26bc48fcc93c8ec36bc0202eeed96a788", size = 5557094, upload-time = "2026-10-22T20:10:15.096Z" },
 ]

 [[package]]
@@ -29,7 +29,19 @@ from langgraph._internal._constants import (
    NS_SEP,
 )

-DEFAULT_RECURSION_LIMIT = int(getenv("LANGGRAPH_DEFAULT_RECURSION_LIMIT", "10007"))
+
+class _DefaultRecursionLimit(int):
+    """The limit `ensure_config` fills in when none is given.
+
+    `merge_configs` skips values of this type rather than any equal value, so an
+    explicit limit that happens to equal the default still applies. It checks the
+    type, not identity, because copying or pickling a config rebuilds the value.
+    """
+
+
+DEFAULT_RECURSION_LIMIT = _DefaultRecursionLimit(
+    getenv("LANGGRAPH_DEFAULT_RECURSION_LIMIT", "10007")
+)
 DELTA_MAX_SUPERSTEPS_SINCE_SNAPSHOT = int(
    getenv("LANGGRAPH_DELTA_MAX_SUPERSTEPS_SINCE_SNAPSHOT", "5000")
 )
@@ -182,7 +194,7 @@ def merge_configs(*configs: RunnableConfig | None) -> RunnableConfig:
                    base.get("callbacks"), cast(Callbacks, value)
                )
            elif key == "recursion_limit":
-                if config["recursion_limit"] != DEFAULT_RECURSION_LIMIT:
+                if not isinstance(config["recursion_limit"], _DefaultRecursionLimit):
                    base["recursion_limit"] = config["recursion_limit"]
            else:
                base[key] = config[key]  # type: ignore[literal-required]
@@ -335,6 +347,7 @@ def ensure_config(*configs: RunnableConfig | None) -> RunnableConfig:
        recursion_limit=DEFAULT_RECURSION_LIMIT,
        configurable={},
    )
+    inherited_tags: set[str] = set()
    if var_config := var_child_runnable_config.get():
        empty.update(
            {
@@ -343,6 +356,7 @@ def ensure_config(*configs: RunnableConfig | None) -> RunnableConfig:
                if _is_not_empty(v)
            },
        )
+        inherited_tags.update(empty["tags"])
    # An explicit config that supplies its own checkpoint coordinate (a
    # thread_id, or any checkpoint_ns/checkpoint_id/checkpoint_map) is addressing
    # its own checkpoint lineage, so drop the inherited ambient configurable
@@ -395,12 +409,14 @@ def ensure_config(*configs: RunnableConfig | None) -> RunnableConfig:
                elif k == "tags":
                    # Concatenate tags across configs so values bound via
                    # with_config(...) are preserved when later configs
-                    # supply additional tags. Matches merge_configs.
+                    # supply additional tags. Matches merge_configs. Tags the
+                    # ambient config already has are skipped: a subgraph gets
+                    # its parent's config both explicitly and as the ambient
+                    # one, so they would double at every nesting level.
                    existing_tags: list[str] | None = empty.get("tags")
+                    new_tags = [t for t in cast(list, v) if t not in inherited_tags]
                    empty["tags"] = (
-                        [*existing_tags, *cast(list, v)]
-                        if existing_tags
-                        else list(cast(list, v))
+                        [*existing_tags, *new_tags] if existing_tags else new_tags
                    )
                else:
                    empty[k] = v  # type: ignore[literal-required]
@@ -1,6 +1,7 @@
 from __future__ import annotations

 import asyncio
+import copy
 import enum
 import inspect
 import logging
@@ -23,9 +24,16 @@ from typing import (
    Optional,
    Protocol,
    TypeGuard,
+    TypeVar,
    cast,
 )

+from langchain_core.callbacks.base import BaseCallbackHandler, BaseCallbackManager
+from langchain_core.callbacks.manager import (
+    AsyncCallbackManagerForChainRun,
+    BaseRunManager,
+    CallbackManagerForChainRun,
+)
 from langchain_core.runnables.base import (
    Runnable,
    RunnableConfig,
@@ -41,6 +49,7 @@ from langchain_core.runnables.config import (
    var_child_runnable_config,
 )
 from langchain_core.runnables.utils import Input, Output
+from langchain_core.tracers.base import BaseTracer
 from langchain_core.tracers.langchain import LangChainTracer
 from langgraph.store.base import BaseStore

@@ -57,7 +66,7 @@ from langgraph._internal._constants import (
 )
 from langgraph._internal._typing import MISSING
 from langgraph.errors import NodeError
-from langgraph.types import StreamWriter
+from langgraph.types import StreamWriter, TracePolicy

 try:
    from langchain_core.tracers._streaming import _StreamingCallbackHandler
@@ -85,6 +94,29 @@ def _trace_payload(value: Any, transform: Callable[[Any], Any] | None) -> Any:
        return value


+def _is_stream_handler(handler: BaseCallbackHandler) -> bool:
+    # Inheritance, not isinstance on the runtime-checkable protocol, which also
+    # matches any exporter that merely has the tap methods: stream handlers get a
+    # node's real payload and its streamed output.
+    return _StreamingCallbackHandler in type(handler).__mro__
+
+
+_M = TypeVar("_M", bound=BaseCallbackManager | BaseRunManager)
+
+
+def _split_handlers(manager: _M) -> tuple[_M, _M]:
+    # The recorded payload exists to keep tracer ingest small, so only tracers get it.
+    # `astream_log`'s handler is a tracer too, but it feeds the caller's own stream.
+    real, tracers = copy.copy(manager), copy.copy(manager)
+    tracers.handlers = [
+        h
+        for h in manager.handlers
+        if isinstance(h, BaseTracer) and not _is_stream_handler(h)
+    ]
+    real.handlers = [h for h in manager.handlers if h not in tracers.handlers]
+    return real, tracers
+
+
 def _set_config_context(
    config: RunnableConfig, run: Any = None
 ) -> Token[RunnableConfig | None]:
@@ -593,13 +625,15 @@ class RunnableSeq(Runnable):
        *steps: RunnableLike,
        name: str | None = None,
        trace_inputs: Callable[[Any], Any] | None = None,
-        trace_outputs: Callable[[Any], Any] | None = None,
+        trace_policy: TracePolicy | None = None,
    ) -> None:
        """Create a new RunnableSeq.

        Args:
            steps: The steps to include in the sequence.
            name: The name of the `Runnable`.
+            trace_inputs: Transforms the input every callback handler receives.
+            trace_policy: Processes what tracers record, after `trace_inputs`.

        Raises:
            ValueError: If the sequence has less than 2 steps.
@@ -619,7 +653,67 @@ class RunnableSeq(Runnable):
        self.steps = steps_flat
        self.name = name
        self.trace_inputs = trace_inputs
-        self.trace_outputs = trace_outputs
+        self.trace_policy = trace_policy or TracePolicy()
+
+    def _start(self, input: Any, config: RunnableConfig) -> CallbackManagerForChainRun:
+        callback_manager = get_callback_manager_for_config(config)
+        payload = _trace_payload(input, self.trace_inputs)
+        recorded = _trace_payload(payload, self.trace_policy.process_inputs)
+        name = config.get("run_name") or self.get_name()
+        run_id = config.pop("run_id", None)
+        if recorded is payload:
+            return callback_manager.on_chain_start(
+                None, payload, name=name, run_id=run_id
+            )
+        real, tracers = _split_handlers(callback_manager)
+        run_manager = real.on_chain_start(None, payload, name=name, run_id=run_id)
+        tracers.on_chain_start(None, recorded, name=name, run_id=run_manager.run_id)
+        run_manager.handlers = callback_manager.handlers
+        return run_manager
+
+    async def _astart(
+        self, input: Any, config: RunnableConfig
+    ) -> AsyncCallbackManagerForChainRun:
+        callback_manager = get_async_callback_manager_for_config(config)
+        payload = _trace_payload(input, self.trace_inputs)
+        recorded = _trace_payload(payload, self.trace_policy.process_inputs)
+        name = config.get("run_name") or self.get_name()
+        run_id = config.pop("run_id", None)
+        if recorded is payload:
+            return await callback_manager.on_chain_start(
+                None, payload, name=name, run_id=run_id
+            )
+        real, tracers = _split_handlers(callback_manager)
+        # Tracers start last, so a cancellation before then leaves no tracer run open.
+        run_manager = await real.on_chain_start(None, payload, name=name, run_id=run_id)
+        await tracers.on_chain_start(
+            None, recorded, name=name, run_id=run_manager.run_id
+        )
+        run_manager.handlers = callback_manager.handlers
+        return run_manager
+
+    def _end(self, run_manager: CallbackManagerForChainRun, output: Any) -> None:
+        recorded = _trace_payload(output, self.trace_policy.process_outputs)
+        if recorded is output:
+            run_manager.on_chain_end(output)
+        else:
+            real, tracers = _split_handlers(run_manager)
+            tracers.on_chain_end(recorded)
+            real.on_chain_end(output)
+
+    async def _aend(
+        self, run_manager: AsyncCallbackManagerForChainRun, output: Any
+    ) -> None:
+        recorded = _trace_payload(output, self.trace_policy.process_outputs)
+        if recorded is output:
+            await run_manager.on_chain_end(output)
+        else:
+            real, tracers = _split_handlers(run_manager)
+            # Tracers end first: on_chain_end is shielded once it starts, so a
+            # cancellation can only skip the other handlers' end, never leave a
+            # tracer run open.
+            await tracers.on_chain_end(recorded)
+            await real.on_chain_end(output)

    def __or__(
        self,
@@ -676,15 +770,8 @@ class RunnableSeq(Runnable):
    ) -> Any:
        if config is None:
            config = ensure_config()
-        # setup callbacks and context
-        callback_manager = get_callback_manager_for_config(config)
        # start the root run
-        run_manager = callback_manager.on_chain_start(
-            None,
-            _trace_payload(input, self.trace_inputs),
-            name=config.get("run_name") or self.get_name(),
-            run_id=config.pop("run_id", None),
-        )
+        run_manager = self._start(input, config)
        # invoke all steps in sequence
        try:
            for i, step in enumerate(self.steps):
@@ -712,7 +799,7 @@ class RunnableSeq(Runnable):
            run_manager.on_chain_error(e)
            raise
        else:
-            run_manager.on_chain_end(_trace_payload(input, self.trace_outputs))
+            self._end(run_manager, input)
            return input

    async def ainvoke(
@@ -723,15 +810,8 @@ class RunnableSeq(Runnable):
    ) -> Any:
        if config is None:
            config = ensure_config()
-        # setup callbacks
-        callback_manager = get_async_callback_manager_for_config(config)
        # start the root run
-        run_manager = await callback_manager.on_chain_start(
-            None,
-            _trace_payload(input, self.trace_inputs),
-            name=config.get("run_name") or self.get_name(),
-            run_id=config.pop("run_id", None),
-        )
+        run_manager = await self._astart(input, config)

        # invoke all steps in sequence
        try:
@@ -765,7 +845,7 @@ class RunnableSeq(Runnable):
            await run_manager.on_chain_error(e)
            raise
        else:
-            await run_manager.on_chain_end(_trace_payload(input, self.trace_outputs))
+            await self._aend(run_manager, input)
            return input

    def stream(
@@ -776,15 +856,8 @@ class RunnableSeq(Runnable):
    ) -> Iterator[Any]:
        if config is None:
            config = ensure_config()
-        # setup callbacks
-        callback_manager = get_callback_manager_for_config(config)
        # start the root run
-        run_manager = callback_manager.on_chain_start(
-            None,
-            _trace_payload(input, self.trace_inputs),
-            name=config.get("run_name") or self.get_name(),
-            run_id=config.pop("run_id", None),
-        )
+        run_manager = self._start(input, config)
        # get the run object
        for h in run_manager.handlers:
            if isinstance(h, LangChainTracer):
@@ -814,10 +887,9 @@ class RunnableSeq(Runnable):
                        )
                        iterator = step.transform(iterator, config)
                # populates streamed_output in astream_log() output if needed
-                if _StreamingCallbackHandler is not None:
-                    for h in run_manager.handlers:
-                        if isinstance(h, _StreamingCallbackHandler):
-                            iterator = h.tap_output_iter(run_manager.run_id, iterator)
+                for h in run_manager.handlers:
+                    if _is_stream_handler(h):
+                        iterator = h.tap_output_iter(run_manager.run_id, iterator)
                # consume into final output
                output = context.run(_consume_iter, iterator)
                # sequence doesn't emit output, yield to mark as generator
@@ -826,7 +898,7 @@ class RunnableSeq(Runnable):
                run_manager.on_chain_error(e)
                raise
            else:
-                run_manager.on_chain_end(_trace_payload(output, self.trace_outputs))
+                self._end(run_manager, output)

    async def astream(
        self,
@@ -836,15 +908,8 @@ class RunnableSeq(Runnable):
    ) -> AsyncIterator[Any]:
        if config is None:
            config = ensure_config()
-        # setup callbacks
-        callback_manager = get_async_callback_manager_for_config(config)
        # start the root run
-        run_manager = await callback_manager.on_chain_start(
-            None,
-            _trace_payload(input, self.trace_inputs),
-            name=config.get("run_name") or self.get_name(),
-            run_id=config.pop("run_id", None),
-        )
+        run_manager = await self._astart(input, config)
        # stream the last steps
        # transform the input stream of each step with the next
        # steps that don't natively support transforming an input stream will
@@ -880,12 +945,11 @@ class RunnableSeq(Runnable):
                            if hasattr(aiterator, "aclose"):
                                stack.push_async_callback(aiterator.aclose)
                        # populates streamed_output in astream_log() output if needed
-                        if _StreamingCallbackHandler is not None:
-                            for h in run_manager.handlers:
-                                if isinstance(h, _StreamingCallbackHandler):
-                                    aiterator = h.tap_output_aiter(
-                                        run_manager.run_id, aiterator
-                                    )
+                        for h in run_manager.handlers:
+                            if _is_stream_handler(h):
+                                aiterator = h.tap_output_aiter(
+                                    run_manager.run_id, aiterator
+                                )
                        # consume into final output
                        output = await asyncio.create_task(
                            _consume_aiter(aiterator), context=context
@@ -896,9 +960,7 @@ class RunnableSeq(Runnable):
                    await run_manager.on_chain_error(e)
                    raise
                else:
-                    await run_manager.on_chain_end(
-                        _trace_payload(output, self.trace_outputs)
-                    )
+                    await self._aend(run_manager, output)
        else:
            try:
                async with AsyncExitStack() as stack:
@@ -914,12 +976,11 @@ class RunnableSeq(Runnable):
                        if hasattr(aiterator, "aclose"):
                            stack.push_async_callback(aiterator.aclose)
                    # populates streamed_output in astream_log() output if needed
-                    if _StreamingCallbackHandler is not None:
-                        for h in run_manager.handlers:
-                            if isinstance(h, _StreamingCallbackHandler):
-                                aiterator = h.tap_output_aiter(
-                                    run_manager.run_id, aiterator
-                                )
+                    for h in run_manager.handlers:
+                        if _is_stream_handler(h):
+                            aiterator = h.tap_output_aiter(
+                                run_manager.run_id, aiterator
+                            )
                    # consume into final output
                    output = await _consume_aiter(aiterator)
                    # sequence doesn't emit output, yield to mark as generator
@@ -928,9 +989,7 @@ class RunnableSeq(Runnable):
                await run_manager.on_chain_error(e)
                raise
            else:
-                await run_manager.on_chain_end(
-                    _trace_payload(output, self.trace_outputs)
-                )
+                await self._aend(run_manager, output)


 def _consume_iter(it: Iterator[Any]) -> Any:
@@ -60,7 +60,7 @@ class DeltaChannel(Generic[Value], BaseChannel[Any, Any, Any]):
        typ: The value type (e.g. `list`, `dict`). Inferred automatically
            from the outer type when used inside `Annotated[T, DeltaChannel(...)]`.
        snapshot_frequency: Every Nth update to this channel writes a snapshot
-            blob (default `1000`). Must be a positive int.
+            blob (default `100`). Must be a positive int.
    """

    __slots__ = ("value", "reducer", "snapshot_frequency")
@@ -71,7 +71,7 @@ class DeltaChannel(Generic[Value], BaseChannel[Any, Any, Any]):
        reducer: Callable[[Any, Sequence[Any]], Any],
        typ: type[Value] | None = None,
        *,
-        snapshot_frequency: int = 1000,
+        snapshot_frequency: int = 100,
    ) -> None:
        if snapshot_frequency <= 0:
            raise ValueError(
@@ -22,7 +22,7 @@ from langchain_core.messages import (
 )
 from typing_extensions import TypedDict, deprecated

-from langgraph._internal._constants import CONF, CONFIG_KEY_SEND, NS_SEP
+from langgraph._internal._constants import CONF, CONFIG_KEY_SEND
 from langgraph.graph.state import StateGraph
 from langgraph.pregel._messages import (
    _messages_delta_reducer as _messages_delta_reducer,
@@ -343,7 +343,7 @@ def push_message(
    )

    from langgraph.config import get_config
-    from langgraph.pregel._messages import StreamMessagesHandler
+    from langgraph.pregel._messages import StreamMessagesHandler, graph_namespace

    config = get_config()
    message = next(x for x in convert_to_messages([message]))
@@ -359,15 +359,12 @@ def push_message(
    ):
        handlers = config["callbacks"]

-    if stream_handler := next(
-        (x for x in handlers if isinstance(x, StreamMessagesHandler)), None
-    ):
-        metadata = config["metadata"]
-        message_meta = (
-            tuple(cast(str, metadata["langgraph_checkpoint_ns"]).split(NS_SEP)),
-            metadata,
-        )
-        stream_handler._emit(message_meta, message, dedupe=False)
+    metadata = config["metadata"]
+    for handler in handlers:
+        if isinstance(handler, StreamMessagesHandler):
+            ns = graph_namespace(metadata)
+            if handler.subgraphs or not ns or ns == handler.parent_ns:
+                handler._emit((ns, metadata), message, dedupe=False)

    if state_key:
        config[CONF][CONFIG_KEY_SEND]([(state_key, message)])
@@ -228,6 +228,10 @@ def draw_graph(
            metadata["__interrupt"] = "before"
        elif name in interrupt_after_nodes:
            metadata["__interrupt"] = "after"
+        # Duck-typed so core doesn't import ToolNode from langgraph-prebuilt.
+        tools_by_name = getattr(node.bound, "tools_by_name", None)
+        if isinstance(tools_by_name, Mapping) and tools_by_name:
+            metadata["__tools"] = list(tools_by_name)
        graph.add_node(node.bound, name, metadata=metadata or None)
    # add start node
    if START not in nodes:
@@ -3,6 +3,7 @@ from __future__ import annotations
 import asyncio
 import concurrent.futures
 import time
+import weakref
 from collections.abc import Awaitable, Callable, Coroutine
 from contextlib import AbstractAsyncContextManager, AbstractContextManager, ExitStack
 from contextvars import copy_context
@@ -37,6 +38,13 @@ class Submit(Protocol[P, T]):
    ) -> concurrent.futures.Future[T]: ...


+# Futures whose exception a node's error handler took over; the executors
+# don't re-raise them on exit.
+SKIP_RERAISE_SET: weakref.WeakSet[concurrent.futures.Future | asyncio.Future] = (
+    weakref.WeakSet()
+)
+
+
 class BackgroundExecutor(AbstractContextManager):
    """A context manager that runs sync tasks in the background.
    Uses a thread pool executor to delegate tasks to separate threads.
@@ -111,7 +119,7 @@ class BackgroundExecutor(AbstractContextManager):
        if exc_type is None:
            # re-raise the first exception that occurred in a task
            for task, (_, reraise) in tasks.items():
-                if not reraise:
+                if not reraise or task in SKIP_RERAISE_SET:
                    continue
                try:
                    task.result()
@@ -202,7 +210,7 @@ class AsyncBackgroundExecutor(AbstractAsyncContextManager):
        if exc_type is None:
            # re-raise the first exception that occurred in a task
            for task, (_, reraise) in tasks.items():
-                if not reraise:
+                if not reraise or task in SKIP_RERAISE_SET:
                    continue
                try:
                    if exc := task.exception():
@@ -132,7 +132,7 @@ from langgraph.pregel.debug import (
    map_debug_tasks,
 )
 from langgraph.pregel.protocol import StreamChunk, StreamProtocol
-from langgraph.runtime import RunControl, Runtime
+from langgraph.runtime import RunControl, Runtime, _RunExits
 from langgraph.types import (
    All,
    CachePolicy,
@@ -657,48 +657,11 @@ class PregelLoop:
        # check if iteration limit is reached
        if self.step > self.stop:
            self.status = "out_of_steps"
+            # stream the checkpoint the last step saved, as a tick that runs would
+            self._prepare_tasks_and_emit_checkpoint()
            return False

-        # prepare next tasks
-        self.tasks = prepare_next_tasks(
-            self.checkpoint,
-            self.checkpoint_pending_writes,
-            self.nodes,
-            self.channels,
-            self.managed,
-            self.config,
-            self.step,
-            self.stop,
-            for_execution=True,
-            manager=self.manager,
-            store=self.store,
-            checkpointer=self.checkpointer,
-            trigger_to_nodes=self.trigger_to_nodes,
-            updated_channels=self.updated_channels,
-            retry_policy=self.retry_policy,
-            cache_policy=self.cache_policy,
-        )
-
-        # produce debug output
-        if self._checkpointer_put_after_previous is not None:
-            self._emit(
-                "checkpoints",
-                map_debug_checkpoint,
-                {
-                    **self.checkpoint_config,
-                    CONF: {
-                        **self.checkpoint_config[CONF],
-                        CONFIG_KEY_CHECKPOINT_ID: self.checkpoint["id"],
-                    },
-                },
-                self.channels,
-                self.stream_keys,
-                self.checkpoint_metadata,
-                self.tasks.values(),
-                self.checkpoint_pending_writes,
-                self.prev_checkpoint_config,
-                self.output_keys,
-            )
+        self._prepare_tasks_and_emit_checkpoint()

        # if no more tasks, we're done
        if not self.tasks:
@@ -806,10 +769,53 @@ class PregelLoop:
            self.checkpoint, self.interrupt_after, self.tasks.values()
        ):
            self.status = "interrupt_after"
+            # the next tick, which streams this checkpoint, never runs
+            self._prepare_tasks_and_emit_checkpoint()
            raise GraphInterrupt()
        # unset resuming flag
        self.config[CONF].pop(CONFIG_KEY_RESUMING, None)

+    def _prepare_tasks_and_emit_checkpoint(self) -> None:
+        self.tasks = prepare_next_tasks(
+            self.checkpoint,
+            self.checkpoint_pending_writes,
+            self.nodes,
+            self.channels,
+            self.managed,
+            self.config,
+            self.step,
+            self.stop,
+            for_execution=True,
+            manager=self.manager,
+            store=self.store,
+            checkpointer=self.checkpointer,
+            trigger_to_nodes=self.trigger_to_nodes,
+            updated_channels=self.updated_channels,
+            retry_policy=self.retry_policy,
+            cache_policy=self.cache_policy,
+        )
+
+        # produce debug output
+        if self._checkpointer_put_after_previous is not None:
+            self._emit(
+                "checkpoints",
+                map_debug_checkpoint,
+                {
+                    **self.checkpoint_config,
+                    CONF: {
+                        **self.checkpoint_config[CONF],
+                        CONFIG_KEY_CHECKPOINT_ID: self.checkpoint["id"],
+                    },
+                },
+                self.channels,
+                self.stream_keys,
+                self.checkpoint_metadata,
+                self.tasks.values(),
+                self.checkpoint_pending_writes,
+                self.prev_checkpoint_config,
+                self.output_keys,
+            )
+
    def match_cached_writes(self) -> Sequence[PregelExecutableTask]:
        raise NotImplementedError

@@ -1491,6 +1497,10 @@ class PregelLoop:
            if self._delta_write_futs is not None:
                self._delta_write_futs.append(fut)

+    def _run_exits(self) -> _RunExits | None:
+        runtime = self.config[CONF].get(CONFIG_KEY_RUNTIME)
+        return runtime._run_exits if isinstance(runtime, Runtime) else None
+
    def _suppress_interrupt(
        self,
        exc_type: type[BaseException] | None,
@@ -1868,6 +1878,10 @@ class SyncPregelLoop(PregelLoop, AbstractContextManager):
        self._exit_delta_writes = (
            [] if self.durability == "exit" and self.checkpointer is not None else None
        )
+        # Bottom of the stack, so run-exit callbacks run last: after the
+        # executor has stopped every task and the final checkpoint is saved.
+        if (exits := self._run_exits()) is not None and exits.bind(is_async=False):
+            self.stack.push(exits.exit)
        self.submit = self.stack.enter_context(BackgroundExecutor(self.config))
        self.channels, self.managed = channels_from_checkpoint(
            self.specs,
@@ -2124,6 +2138,10 @@ class AsyncPregelLoop(PregelLoop, AbstractAsyncContextManager):
        self._exit_delta_writes = (
            [] if self.durability == "exit" and self.checkpointer is not None else None
        )
+        # Bottom of the stack, so run-exit callbacks run last: after the
+        # executor has stopped every task and the final checkpoint is saved.
+        if (exits := self._run_exits()) is not None and exits.bind(is_async=True):
+            self.stack.push_async_exit(exits.aexit)
        self.submit = await self.stack.enter_async_context(
            AsyncBackgroundExecutor(self.config)
        )
@@ -23,7 +23,7 @@ from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, LLMResul
 from pydantic import BaseModel

 from langgraph._internal._config import filter_to_user_tags
-from langgraph._internal._constants import NS_SEP
+from langgraph._internal._constants import NS_SEP, PUSH
 from langgraph.channels.binop import _get_overwrite
 from langgraph.channels.delta import DeltaChannel
 from langgraph.constants import TAG_HIDDEN, TAG_NOSTREAM
@@ -55,6 +55,23 @@ def _state_values(obj: Any) -> Sequence[Any]:
    return ()


+def graph_namespace(metadata: Mapping[str, Any]) -> tuple[str, ...]:
+    """Namespace of the graph a task runs in, from the task's run metadata.
+
+    Drops the task's own segment, plus one per enclosing `@task` call: a call
+    runs in its caller's graph but nests its segment under the caller's. A
+    call's `langgraph_path` is `(PUSH, <caller's path>, <index>, True)`, while
+    a `Send` task's ends in `False`.
+    """
+    ns = tuple(cast(str, metadata["langgraph_checkpoint_ns"]).split(NS_SEP))
+    path = metadata.get("langgraph_path", ())
+    drop = 1
+    while len(path) == 4 and path[0] == PUSH and path[-1] is True:
+        drop += 1
+        path = path[1]
+    return ns[:-drop]
+
+
 class StreamMessagesHandler(BaseCallbackHandler, _StreamingCallbackHandler):
    """A callback handler that implements stream_mode=messages.

@@ -148,9 +165,7 @@ class StreamMessagesHandler(BaseCallbackHandler, _StreamingCallbackHandler):
        **kwargs: Any,
    ) -> Any:
        if metadata and (not tags or (TAG_NOSTREAM not in tags)):
-            ns = tuple(cast(str, metadata["langgraph_checkpoint_ns"]).split(NS_SEP))[
-                :-1
-            ]
+            ns = graph_namespace(metadata)
            if not self.subgraphs and len(ns) > 0 and ns != self.parent_ns:
                return
            if (filtered_tags := filter_to_user_tags(tags)) is not None:
@@ -213,10 +228,8 @@ class StreamMessagesHandler(BaseCallbackHandler, _StreamingCallbackHandler):
            and kwargs.get("name") == metadata.get("langgraph_node")
            and (not tags or TAG_HIDDEN not in tags)
        ):
-            ns = tuple(cast(str, metadata["langgraph_checkpoint_ns"]).split(NS_SEP))[
-                :-1
-            ]
-            if not self.subgraphs and len(ns) > 0:
+            ns = graph_namespace(metadata)
+            if not self.subgraphs and len(ns) > 0 and ns != self.parent_ns:
                return
            self.metadata[run_id] = (ns, metadata)
            for value in _state_values(inputs):
@@ -227,23 +227,14 @@ class PregelNode:
    def node(self) -> Runnable[Any, Any] | None:
        """Get a runnable that combines `bound` and `writers`."""
        writers = self.flat_writers
-        trace_inputs = self.trace_policy.process_inputs if self.trace_policy else None
-        trace_outputs = self.trace_policy.process_outputs if self.trace_policy else None
        if self.bound is DEFAULT_BOUND and not writers:
            return None
        elif self.bound is DEFAULT_BOUND and len(writers) == 1:
            return writers[0]
        elif self.bound is DEFAULT_BOUND:
-            return RunnableSeq(
-                *writers, trace_inputs=trace_inputs, trace_outputs=trace_outputs
-            )
+            return RunnableSeq(*writers, trace_policy=self.trace_policy)
        elif writers:
-            return RunnableSeq(
-                self.bound,
-                *writers,
-                trace_inputs=trace_inputs,
-                trace_outputs=trace_outputs,
-            )
+            return RunnableSeq(self.bound, *writers, trace_policy=self.trace_policy)
        else:
            return self.bound

@@ -34,6 +34,7 @@ from langgraph._internal._constants import (
    CONFIG_KEY_THREAD_ID,
    CONFIG_KEY_TIMED_ATTEMPT_OBSERVER,
    NS_SEP,
+    RETURN,
 )
 from langgraph._internal._runnable import create_task_in_config_context
 from langgraph._internal._timeout import sync_timeout_unsupported
@@ -733,15 +734,14 @@ async def arun_with_retry(
                if stream:
                    async for _ in task.proc.astream(task.input, config):
                        pass
-                    break
+                    return _return_value(task)
                return await task.proc.ainvoke(task.input, config)
            result = await _arun_with_timeout(
                task, config, resolved_timeout, attempt_ctx, stream=stream
            )
            _finish_timed_attempt(config, attempt_ctx)
            if stream:
-                # if successful, end
-                break
+                return _return_value(task)
            return result
        except ParentCommand as exc:
            ns: str = config[CONF][CONFIG_KEY_CHECKPOINT_NS]
@@ -831,6 +831,12 @@ async def arun_with_retry(
            config = patch_configurable(config, {CONFIG_KEY_RESUMING: True})


+def _return_value(task: PregelExecutableTask) -> Any:
+    # Streaming drains the task instead of returning its result, so a caller
+    # awaiting a `@task` needs the value the task wrote.
+    return next((value for channel, value in task.writes if channel == RETURN), None)
+
+
 def _should_retry_on(retry_policy: RetryPolicy, exc: Exception) -> bool:
    """Check if the given exception should be retried based on the retry policy."""
    if isinstance(retry_policy.retry_on, Sequence):
@@ -43,7 +43,7 @@ from langgraph._internal._typing import MISSING
 from langgraph.constants import TAG_HIDDEN
 from langgraph.errors import GraphBubbleUp, GraphInterrupt, is_invalid_resume
 from langgraph.pregel._algo import Call
-from langgraph.pregel._executor import Submit
+from langgraph.pregel._executor import SKIP_RERAISE_SET, Submit
 from langgraph.pregel._retry import arun_with_retry, run_with_retry
 from langgraph.pregel._task_status import CONTROL_WRITES
 from langgraph.types import (
@@ -68,10 +68,6 @@ EXCLUDED_FRAME_FNAMES = (
    "concurrent/futures/_base.py",
 )

-SKIP_RERAISE_SET: weakref.WeakSet[concurrent.futures.Future | asyncio.Future] = (
-    weakref.WeakSet()
-)
-

 class FuturesDict(Generic[F, E], dict[F, PregelExecutableTask | None]):
    event: E
@@ -239,6 +235,8 @@ class PregelRunner:
                        fut.set_exception(exc)
                        futures.done.add(fut)
                elif reraise and id(exc) not in self._handled_exception_ids:
+                    # let the caller stream this task's result before it raises
+                    yield
                    if tb := exc.__traceback__:
                        while tb.tb_next is not None and any(
                            tb.tb_frame.f_code.co_filename.endswith(name)
@@ -430,6 +428,8 @@ class PregelRunner:
                        fut.set_exception(exc)
                        futures.done.add(fut)
                elif reraise and id(exc) not in self._handled_exception_ids:
+                    # let the caller stream this task's result before it raises
+                    yield
                    if tb := exc.__traceback__:
                        while tb.tb_next is not None and any(
                            tb.tb_frame.f_code.co_filename.endswith(name)
@@ -7,8 +7,9 @@ from uuid import UUID

 from langchain_core.callbacks import BaseCallbackHandler

-from langgraph._internal._constants import NS_SEP
 from langgraph.constants import TAG_NOSTREAM
+from langgraph.errors import GraphDrained, GraphInterrupt
+from langgraph.pregel._messages import graph_namespace
 from langgraph.pregel.protocol import StreamChunk

 try:
@@ -91,12 +92,9 @@ class StreamToolCallHandler(BaseCallbackHandler, _StreamingCallbackHandler):
    ) -> tuple[str, ...] | None:
        """Resolve the namespace this tool call should emit at, or `None` to skip.

-        Mirrors `StreamMessagesHandler.on_chat_model_start`'s namespace
-        derivation: parses `langgraph_checkpoint_ns` (which ends with
-        the `node_name:task_id` of the calling node), drops that
-        trailing segment, and returns the containing subgraph's own
-        namespace. Returns `None` when the call should be silently
-        suppressed:
+        Mirrors `StreamMessagesHandler.on_chat_model_start`: the namespace
+        is that of the graph the tool runs in (`graph_namespace`). Returns
+        `None` when the call should be silently suppressed:

        - `metadata` is missing — handler is attached to a context
          without Pregel routing info.
@@ -109,11 +107,9 @@ class StreamToolCallHandler(BaseCallbackHandler, _StreamingCallbackHandler):
            return None
        if tags and TAG_NOSTREAM in tags:
            return None
-        nskey = metadata.get("langgraph_checkpoint_ns")
-        if not nskey:
-            ns: tuple[str, ...] = ()
-        else:
-            ns = tuple(cast(str, nskey).split(NS_SEP))[:-1]
+        ns = (
+            graph_namespace(metadata) if metadata.get("langgraph_checkpoint_ns") else ()
+        )
        if not self.subgraphs and len(ns) > 0 and ns != self.parent_ns:
            return None
        return ns
@@ -188,6 +184,10 @@ class StreamToolCallHandler(BaseCallbackHandler, _StreamingCallbackHandler):
            return
        ns, tool_call_id, token = info
        self._reset_writer(token)
+        # An interrupted or drained tool is paused, not failed: it runs again
+        # on resume and reports its start and end then.
+        if isinstance(error, (GraphInterrupt, GraphDrained)):
+            return
        self.stream(
            (
                ns,
@@ -97,6 +97,7 @@ from langgraph._internal._runnable import (
    Runnable,
    RunnableLike,
    RunnableSeq,
+    _is_stream_handler,
    coerce_to_runnable,
 )
 from langgraph._internal._timeout import coerce_timeout_policy
@@ -174,6 +175,7 @@ from langgraph.runtime import (
    RunControl,
    Runtime,
    ServerInfo,
+    _RunExits,
 )
 from langgraph.stream._mux import StreamMux
 from langgraph.stream._types import StreamTransformer
@@ -203,11 +205,6 @@ from langgraph.types import (
 from langgraph.typing import ContextT, InputT, OutputT, StateT
 from langgraph.warnings import LangGraphDeprecatedSinceV10

-try:
-    from langchain_core.tracers._streaming import _StreamingCallbackHandler
-except ImportError:
-    _StreamingCallbackHandler = None  # type: ignore
-
 __all__ = ("NodeBuilder", "Pregel")

 _WriteValue = Callable[[Input], Output] | Any
@@ -2880,6 +2877,20 @@ class Pregel(
                    for h in callback_manager.inheritable_handlers
                    if not isinstance(h, StreamMessagesHandlerV2)
                ]
+            if CONFIG_KEY_TASK_ID not in config[CONF]:
+                # Not run as a task (a child invoked with its own thread, say),
+                # this graph's namespace starts at the root, so a caller's
+                # handler would take this graph's events for its own graph's.
+                callback_manager.handlers = [
+                    h
+                    for h in callback_manager.handlers
+                    if not _is_stream_mode_handler(h)
+                ]
+                callback_manager.inheritable_handlers = [
+                    h
+                    for h in callback_manager.inheritable_handlers
+                    if not _is_stream_mode_handler(h)
+                ]
            if "ls_integration" not in callback_manager.metadata:
                callback_manager.add_metadata({"ls_integration": "langgraph"})
            run_manager = callback_manager.on_chain_start(
@@ -2898,9 +2909,9 @@ class Pregel(
                )
            # set up subgraph checkpointing
            config = self._own_checkpoint_config(config)
+            stream_handlers: list[StreamMessagesHandler | StreamToolCallHandler] = []
            # set up messages stream mode
            if "messages" in stream_modes:
-                ns_ = cast(str | None, config[CONF].get(CONFIG_KEY_CHECKPOINT_NS))
                use_stream_messages_v2 = bool(
                    version == "v2" and config[CONF].get(CONFIG_KEY_STREAM_MESSAGES_V2)
                )
@@ -2909,24 +2920,12 @@ class Pregel(
                    if use_stream_messages_v2
                    else StreamMessagesHandler
                )
-                run_manager.inheritable_handlers.append(
-                    messages_handler_cls(
-                        stream.put,
-                        subgraphs,
-                        parent_ns=tuple(ns_.split(NS_SEP)) if ns_ else None,
-                    )
-                )
+                stream_handlers.append(messages_handler_cls(stream.put, subgraphs))

            # set up tools stream mode
            if "tools" in stream_modes:
-                ns_tools = cast(str | None, config[CONF].get(CONFIG_KEY_CHECKPOINT_NS))
-                run_manager.inheritable_handlers.append(
-                    StreamToolCallHandler(
-                        stream.put,
-                        subgraphs,
-                        parent_ns=tuple(ns_tools.split(NS_SEP)) if ns_tools else None,
-                    )
-                )
+                stream_handlers.append(StreamToolCallHandler(stream.put, subgraphs))
+            run_manager.inheritable_handlers.extend(stream_handlers)

            # set up custom stream mode
            if "custom" in stream_modes:
@@ -2968,6 +2967,8 @@ class Pregel(
                execution_info=None,
                server_info=server_info,
                control=control or parent_runtime.control or RunControl(),
+                # a subgraph shares the root run's exit callbacks
+                _run_exits=parent_runtime._run_exits or _RunExits(),
            )
            runtime = parent_runtime.merge(runtime)
            config[CONF][CONFIG_KEY_RUNTIME] = runtime
@@ -3009,6 +3010,10 @@ class Pregel(
                cache_policy=self.cache_policy,
                has_graph_lifecycle_callbacks=bool(graph_callback_manager.handlers),
            ) as loop:
+                # Known only now: the loop numbers the namespace of every
+                # subgraph a task runs after its first.
+                for handler in stream_handlers:
+                    handler.parent_ns = loop.checkpoint_ns
                emit_graph_lifecycle_events(loop)
                # create runner
                runner = PregelRunner(
@@ -3291,6 +3296,20 @@ class Pregel(
                    for h in callback_manager.inheritable_handlers
                    if not isinstance(h, StreamMessagesHandlerV2)
                ]
+            if CONFIG_KEY_TASK_ID not in config[CONF]:
+                # Not run as a task (a child invoked with its own thread, say),
+                # this graph's namespace starts at the root, so a caller's
+                # handler would take this graph's events for its own graph's.
+                callback_manager.handlers = [
+                    h
+                    for h in callback_manager.handlers
+                    if not _is_stream_mode_handler(h)
+                ]
+                callback_manager.inheritable_handlers = [
+                    h
+                    for h in callback_manager.inheritable_handlers
+                    if not _is_stream_mode_handler(h)
+                ]
            if "ls_integration" not in callback_manager.metadata:
                callback_manager.add_metadata({"ls_integration": "langgraph"})
            run_manager = await callback_manager.on_chain_start(
@@ -3304,18 +3323,10 @@ class Pregel(
                run_id=run_manager.run_id,
            )
            # if running from astream_log() run each proc with streaming
-            do_stream = (
-                next(
-                    (
-                        True
-                        for h in run_manager.handlers
-                        if isinstance(h, _StreamingCallbackHandler)
-                        and not isinstance(h, StreamMessagesHandler)
-                    ),
-                    False,
-                )
-                if _StreamingCallbackHandler is not None
-                else False
+            do_stream = any(
+                _is_stream_handler(h)
+                and not isinstance(h, (StreamMessagesHandler, StreamToolCallHandler))
+                for h in run_manager.handlers
            )
            if checkpointer is None and durability is not None:
                warnings.warn(
@@ -3323,10 +3334,9 @@ class Pregel(
                )
            # set up subgraph checkpointing
            config = self._own_checkpoint_config(config)
+            stream_handlers: list[StreamMessagesHandler | StreamToolCallHandler] = []
            # set up messages stream mode
            if "messages" in stream_modes:
-                # namespace can be None in a root level graph?
-                ns_ = cast(str | None, config[CONF].get(CONFIG_KEY_CHECKPOINT_NS))
                use_stream_messages_v2 = bool(
                    version == "v2" and config[CONF].get(CONFIG_KEY_STREAM_MESSAGES_V2)
                )
@@ -3335,24 +3345,12 @@ class Pregel(
                    if use_stream_messages_v2
                    else StreamMessagesHandler
                )
-                run_manager.inheritable_handlers.append(
-                    messages_handler_cls(
-                        stream_put,
-                        subgraphs,
-                        parent_ns=tuple(ns_.split(NS_SEP)) if ns_ else None,
-                    )
-                )
+                stream_handlers.append(messages_handler_cls(stream_put, subgraphs))

            # set up tools stream mode
            if "tools" in stream_modes:
-                ns_tools = cast(str | None, config[CONF].get(CONFIG_KEY_CHECKPOINT_NS))
-                run_manager.inheritable_handlers.append(
-                    StreamToolCallHandler(
-                        stream_put,
-                        subgraphs,
-                        parent_ns=tuple(ns_tools.split(NS_SEP)) if ns_tools else None,
-                    )
-                )
+                stream_handlers.append(StreamToolCallHandler(stream_put, subgraphs))
+            run_manager.inheritable_handlers.extend(stream_handlers)

            # set up custom stream mode
            def stream_writer(c: Any) -> None:
@@ -3409,6 +3407,8 @@ class Pregel(
                execution_info=None,
                server_info=server_info,
                control=control or parent_runtime.control or RunControl(),
+                # a subgraph shares the root run's exit callbacks
+                _run_exits=parent_runtime._run_exits or _RunExits(),
            )
            runtime = parent_runtime.merge(runtime)
            config[CONF][CONFIG_KEY_RUNTIME] = runtime
@@ -3461,6 +3461,10 @@ class Pregel(
                cache_policy=self.cache_policy,
                has_graph_lifecycle_callbacks=bool(graph_callback_manager.handlers),
            ) as loop:
+                # Known only now: the loop numbers the namespace of every
+                # subgraph a task runs after its first.
+                for handler in stream_handlers:
+                    handler.parent_ns = loop.checkpoint_ns
                await aemit_graph_lifecycle_events(loop)
                # create runner
                runner = PregelRunner(
@@ -4458,6 +4462,10 @@ def _coerce_checkpoint_values(payload: Any, mapper: Callable[[Any], Any]) -> Non
        payload["values"] = mapper(payload["values"])


+def _is_stream_mode_handler(handler: object) -> bool:
+    return isinstance(handler, (StreamMessagesHandler, StreamToolCallHandler))
+
+
 def _resolve_parent_ns(
    graph_config: RunnableConfig | None, call_config: RunnableConfig | None
 ) -> tuple[str, ...]:
@@ -847,6 +847,11 @@ class RemoteGraph(PregelProtocol):
            # filter for what was actually requested
            if mode not in requested:
                continue
+            # subgraph events may have been requested only for the parent stream.
+            # messages are exempt: the server dedupes them across namespaces, so
+            # dropping them here would also drop the root-level message
+            if not subgraphs and NS_SEP in chunk.event and mode != "messages":
+                continue

            if chunk.event.startswith("messages"):
                chunk = chunk._replace(data=tuple(chunk.data))
@@ -861,11 +866,6 @@ class RemoteGraph(PregelProtocol):
                    )
                yield {"type": mode, "ns": ns, "data": chunk.data, "interrupts": ints}
            elif subgraphs:
-                if NS_SEP in chunk.event:
-                    mode, ns_ = chunk.event.split(NS_SEP, 1)
-                    ns = tuple(ns_.split(NS_SEP))
-                else:
-                    mode, ns = chunk.event, ()
                if req_single:
                    yield ns, chunk.data
                else:
@@ -1002,6 +1002,11 @@ class RemoteGraph(PregelProtocol):
            # filter for what was actually requested
            if mode not in requested:
                continue
+            # subgraph events may have been requested only for the parent stream.
+            # messages are exempt: the server dedupes them across namespaces, so
+            # dropping them here would also drop the root-level message
+            if not subgraphs and NS_SEP in chunk.event and mode != "messages":
+                continue

            if chunk.event.startswith("messages"):
                chunk = chunk._replace(data=tuple(chunk.data))
@@ -1016,11 +1021,6 @@ class RemoteGraph(PregelProtocol):
                    )
                yield {"type": mode, "ns": ns, "data": chunk.data, "interrupts": ints}
            elif subgraphs:
-                if NS_SEP in chunk.event:
-                    mode, ns_ = chunk.event.split(NS_SEP, 1)
-                    ns = tuple(ns_.split(NS_SEP))
-                else:
-                    mode, ns = chunk.event, ()
                if req_single:
                    yield ns, chunk.data
                else:
@@ -1,7 +1,11 @@
 from __future__ import annotations

-from collections.abc import Callable
+import inspect
+import logging
+import threading
+from collections.abc import Awaitable, Callable
 from dataclasses import dataclass, field, replace
+from types import TracebackType
 from typing import Any, Generic, cast

 from langgraph.store.base import BaseStore
@@ -104,6 +108,129 @@ class RunControl:
        return self._drain_reason


+logger = logging.getLogger(__name__)
+
+ExitCallback = Callable[[], Awaitable[None] | None]
+"""A cleanup callback registered with `Runtime.on_exit`."""
+
+
+class _RunExits:
+    """Cleanup callbacks for one root graph invocation.
+
+    Created by the root `stream`/`astream` call and shared by every subgraph
+    it runs, through `Runtime`. The root Pregel loop binds it at the bottom of
+    its exit stack, so callbacks run after the loop has stopped every task,
+    persisted its final checkpoint and handled any interrupt, whatever the
+    reason the run ended.
+    """
+
+    __slots__ = ("_callbacks", "_is_async", "_lock", "_state")
+
+    def __init__(self) -> None:
+        self._callbacks: list[ExitCallback] = []
+        self._lock = threading.Lock()
+        self._is_async = False
+        # "new" -> "bound" (owned by a loop) -> "closed" (callbacks ran)
+        self._state = "new"
+
+    def bind(self, *, is_async: bool) -> bool:
+        """Claim these exits for a loop. Only the first (root) loop wins."""
+        with self._lock:
+            if self._state != "new":
+                return False
+            self._state = "bound"
+            self._is_async = is_async
+            return True
+
+    def register(self, callback: ExitCallback) -> None:
+        if not callable(callback):
+            msg = "on_exit() expects a callable"
+            raise TypeError(msg)
+        with self._lock:
+            if self._state == "closed":
+                msg = "on_exit() called after the graph run has already exited"
+                raise RuntimeError(msg)
+            if not self._is_async and _is_async_callable(callback):
+                msg = (
+                    "on_exit() got an async callback in a synchronous graph run "
+                    "(invoke/stream). Register a sync callback, or run the graph "
+                    "with ainvoke/astream."
+                )
+                raise TypeError(msg)
+            self._callbacks.append(callback)
+
+    def _drain(self) -> list[ExitCallback]:
+        with self._lock:
+            self._state = "closed"
+            callbacks = self._callbacks[::-1]  # LIFO
+            self._callbacks = []
+        return callbacks
+
+    def exit(
+        self,
+        exc_type: type[BaseException] | None,
+        exc_value: BaseException | None,
+        traceback: TracebackType | None,
+    ) -> bool:
+        errors: list[Exception] = []
+        for callback in self._drain():
+            try:
+                result = callback()
+                if inspect.isawaitable(result):
+                    if inspect.iscoroutine(result):
+                        result.close()
+                    msg = (
+                        f"on_exit() callback {callback!r} returned an awaitable "
+                        "in a synchronous graph run; it was not awaited."
+                    )
+                    raise TypeError(msg)
+            except Exception as e:
+                errors.append(e)
+        return self._finish(errors, exc_value)
+
+    async def aexit(
+        self,
+        exc_type: type[BaseException] | None,
+        exc_value: BaseException | None,
+        traceback: TracebackType | None,
+    ) -> bool:
+        errors: list[Exception] = []
+        for callback in self._drain():
+            try:
+                result = callback()
+                if inspect.isawaitable(result):
+                    await result
+            except Exception as e:
+                errors.append(e)
+        return self._finish(errors, exc_value)
+
+    @staticmethod
+    def _finish(errors: list[Exception], exc_value: BaseException | None) -> bool:
+        # Every callback runs even if an earlier one failed. A failing callback
+        # never replaces the run's own exception (error, cancellation, stream
+        # close): it is logged instead. If the run otherwise ended cleanly, the
+        # first callback error is raised so cleanup bugs aren't silent.
+        if not errors:
+            return False
+        if exc_value is None:
+            for e in errors[1:]:
+                logger.error("on_exit() callback failed", exc_info=e)
+            raise errors[0]
+        for e in errors:
+            logger.error(
+                "on_exit() callback failed while the graph run was exiting with %r",
+                exc_value,
+                exc_info=e,
+            )
+        return False
+
+
+def _is_async_callable(callback: Any) -> bool:
+    return inspect.iscoroutinefunction(callback) or inspect.iscoroutinefunction(
+        getattr(callback, "__call__", None)
+    )
+
+
 def _no_op_stream_writer(_: Any) -> None: ...


@@ -237,6 +364,40 @@ class Runtime(Generic[ContextT]):
    graph runtime.
    """

+    _run_exits: _RunExits | None = field(default=None, repr=False, compare=False)
+    """Exit callbacks of the root graph run. Use `on_exit` to register one."""
+
+    def on_exit(self, callback: ExitCallback) -> None:
+        """Register cleanup to run when the current graph run ends.
+
+        The callback runs once the root invocation (the `invoke`/`stream`/
+        `ainvoke`/`astream` call the caller made) exits for any reason: it
+        completed, raised, was cancelled, paused on an interrupt, or the caller
+        closed the stream or broke out of it. Calls from subgraphs register on
+        that same root run.
+
+        Callbacks take no arguments and run in reverse registration order,
+        after every task of the run has stopped and the final checkpoint has
+        been written. In `ainvoke`/`astream` a callback may be sync or async; in
+        `invoke`/`stream` it must be sync. A failing callback never hides the
+        run's own exception: it is logged, and raised only if the run otherwise
+        ended without one.
+
+        A process that is killed does not run its callbacks, so use them to
+        free resources early, not as the only guarantee they are freed.
+
+        Args:
+            callback: A zero-argument sync or async callable.
+
+        Raises:
+            RuntimeError: When called outside a graph run, or after it exited.
+            TypeError: When given an async callback in a synchronous run.
+        """
+        if self._run_exits is None:
+            msg = "on_exit() can only be called during a graph run"
+            raise RuntimeError(msg)
+        self._run_exits.register(callback)
+
    def merge(self, other: Runtime[ContextT]) -> Runtime[ContextT]:
        """Merge two runtimes together.

@@ -255,6 +416,7 @@ class Runtime(Generic[ContextT]):
            execution_info=other.execution_info or self.execution_info,
            server_info=other.server_info or self.server_info,
            control=other.control or self.control,
+            _run_exits=other._run_exits or self._run_exits,
        )

    def override(
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"

 [project]
 name = "langgraph"
-version = "1.2.16"
+version = "1.2.19"
 description = "Building stateful, multi-actor applications with LLMs"
 authors = []
 requires-python = ">=3.10"
@@ -133,6 +133,11 @@
            "ToolNode"
          ],
          "name": "tools"
+        },
+        "metadata": {
+          "__tools": [
+            "search_api"
+          ]
        }
      },
      {
@@ -221,6 +226,11 @@
            "ToolNode"
          ],
          "name": "tools"
+        },
+        "metadata": {
+          "__tools": [
+            "search_api"
+          ]
        }
      },
      {
@@ -20,6 +20,7 @@ from langgraph.errors import EmptyChannelError, InvalidUpdateError
 from langgraph.graph import START, StateGraph
 from langgraph.graph.message import _messages_delta_reducer
 from langgraph.graph.state import _get_channel
+from langgraph.pregel._checkpoint import delta_channels_to_snapshot
 from langgraph.types import Overwrite

 pytestmark = pytest.mark.anyio
@@ -342,6 +343,26 @@ def test_delta_channel_checkpoint_returns_missing() -> None:
 # ---------------------------------------------------------------------------


+@pytest.mark.parametrize("frequency", [None, 50, 1000])
+def test_delta_channel_snapshot_frequency(frequency: int | None) -> None:
+    spec = (
+        DeltaChannel(_messages_delta_reducer)
+        if frequency is None
+        else DeltaChannel(_messages_delta_reducer, snapshot_frequency=frequency)
+    )
+    expected = 100 if frequency is None else frequency
+    assert spec.snapshot_frequency == expected
+    channel = spec.from_checkpoint(MISSING)
+    channel.update([HumanMessage(content="hi", id="h1")])
+    assert channel.copy().snapshot_frequency == expected
+    for updates in (expected - 1, expected):
+        assert delta_channels_to_snapshot(
+            {"messages": channel},
+            {"messages": (updates, updates)},
+            {"messages": 1},
+        ) == ({"messages"} if updates == expected else set())
+
+
 def test_delta_channel_snapshot_version_based() -> None:
    """Snapshots fire when a channel accumulates `snapshot_frequency` updates.

@@ -1,11 +1,19 @@
+from collections.abc import Callable
+
 import pytest
 from langchain_core.callbacks import AsyncCallbackManager, BaseCallbackHandler

-from langgraph._internal._config import get_async_callback_manager_for_config
+from langgraph._internal._config import (
+    DEFAULT_RECURSION_LIMIT,
+    get_async_callback_manager_for_config,
+)
 from langgraph.graph import StateGraph
+from langgraph.pregel import Pregel

 pytestmark = pytest.mark.anyio

+EXPLICIT_DEFAULT_LIMIT = int(DEFAULT_RECURSION_LIMIT)
+

 def test_new_async_manager_includes_tags() -> None:
    config = {"callbacks": None}
@@ -114,3 +122,27 @@ async def test_with_config_tags_preserved_on_invoke() -> None:
    await graph.ainvoke({}, {"tags": ["invoke"]})
    assert "bound" in captured, "bound tag was dropped by ensure_config overwrite"
    assert "invoke" in captured, "invoke-time tag not present"
+
+
+@pytest.mark.parametrize(
+    "rebind",
+    [
+        lambda g: g.with_config(recursion_limit=EXPLICIT_DEFAULT_LIMIT),
+        lambda g: g.with_config({"recursion_limit": EXPLICIT_DEFAULT_LIMIT}),
+        lambda g: g.with_config(
+            {"recursion_limit": 1}, recursion_limit=EXPLICIT_DEFAULT_LIMIT
+        ),
+        lambda g: g.with_config(recursion_limit=EXPLICIT_DEFAULT_LIMIT).with_config(
+            tags=["t"]
+        ),
+    ],
+    ids=["kwarg", "config", "kwarg_over_config", "kept_by_later_with_config"],
+)
+def test_with_config_applies_recursion_limit_equal_to_default(
+    rebind: Callable[[Pregel], Pregel],
+) -> None:
+    builder = StateGraph(dict)
+    builder.add_node("node", lambda state: state)
+    builder.add_edge("__start__", "node")
+    graph = builder.compile().with_config(recursion_limit=EXPLICIT_DEFAULT_LIMIT + 1)
+    assert rebind(graph).config["recursion_limit"] == EXPLICIT_DEFAULT_LIMIT
@@ -4346,6 +4346,25 @@ def test_enum_node_names():
    assert graph.invoke({"foo": "hello"}) == {"foo": "hello", "bar": "hello!"}


+@pytest.mark.parametrize("stream_mode", ["tasks", "debug"])
+def test_failing_task_result_is_streamed_before_the_error(stream_mode: str) -> None:
+    class State(TypedDict):
+        x: str
+
+    def failing(state: State) -> State:
+        raise ValueError("boom")
+
+    graph = StateGraph(State).add_node("failing", failing).add_edge(START, "failing")
+    errors = []
+    with pytest.raises(ValueError, match="boom"):
+        for chunk in graph.compile().stream({"x": ""}, stream_mode=stream_mode):
+            payload = chunk["payload"] if stream_mode == "debug" else chunk
+            if "error" in payload:
+                errors.append(payload["error"])
+
+    assert [type(error) for error in errors] == [ValueError]
+
+
 def test_debug_retry(sync_checkpointer: BaseCheckpointSaver):
    class State(TypedDict):
        messages: Annotated[list[str], operator.add]
@@ -4410,6 +4429,52 @@ def test_debug_retry(sync_checkpointer: BaseCheckpointSaver):
        assert stream_parent_conf == history_parent_conf


+def test_interrupt_after_streams_the_checkpoint_it_pauses_on(
+    sync_checkpointer: BaseCheckpointSaver,
+) -> None:
+    class State(TypedDict):
+        x: int
+
+    graph = (
+        StateGraph(State)
+        .add_node("one", lambda state: {"x": 1})
+        .add_node("two", lambda state: {"x": 2})
+        .add_edge(START, "one")
+        .add_edge("one", "two")
+        .compile(checkpointer=sync_checkpointer, interrupt_after=["one"])
+    )
+    config = {"configurable": {"thread_id": "1"}}
+
+    checkpoints = list(graph.stream({"x": 0}, config, stream_mode="checkpoints"))
+
+    state = graph.get_state(config)
+    assert checkpoints[-1]["config"] == state.config
+    assert checkpoints[-1]["next"] == ["two"]
+
+
+def test_recursion_limit_streams_the_last_checkpoint_saved(
+    sync_checkpointer: BaseCheckpointSaver,
+) -> None:
+    class State(TypedDict):
+        x: int
+
+    graph = (
+        StateGraph(State)
+        .add_node("loop", lambda state: {"x": state["x"] + 1})
+        .add_edge(START, "loop")
+        .add_edge("loop", "loop")
+        .compile(checkpointer=sync_checkpointer)
+    )
+    config = {"configurable": {"thread_id": "1"}, "recursion_limit": 3}
+    checkpoints = []
+
+    with pytest.raises(GraphRecursionError):
+        for chunk in graph.stream({"x": 0}, config, stream_mode="checkpoints"):
+            checkpoints.append(chunk)
+
+    assert checkpoints[-1]["config"] == graph.get_state(config).config
+
+
 def test_debug_subgraphs(
    sync_checkpointer: BaseCheckpointSaver, durability: Durability
 ):
@@ -6880,6 +6945,22 @@ def test_named_tasks_functional() -> None:
    ]


+def test_invoke_tags_reach_a_node_three_subgraphs_deep_once() -> None:
+    class State(TypedDict):
+        copies: int
+
+    def leaf(state: State, config: RunnableConfig) -> State:
+        return {"copies": config["tags"].count("my-tag")}
+
+    graph = StateGraph(State).add_node("leaf", leaf).add_edge(START, "leaf").compile()
+    for _ in range(3):
+        graph = (
+            StateGraph(State).add_node("sub", graph).add_edge(START, "sub").compile()
+        )
+
+    assert graph.invoke({"copies": 0}, {"tags": ["my-tag"]}) == {"copies": 1}
+
+
 def test_tags_stream_mode_messages() -> None:
    model = GenericFakeChatModel(messages=iter(["foo"]), tags=["meow"])
    graph = (
@@ -8869,6 +8950,39 @@ def test_subgraph_streaming_sync() -> None:
    assert result["num_chunks"] == 9


+def test_subgraph_streaming_node_output_messages_sync() -> None:
+    """Messages returned by a node (not streamed by an LLM) are emitted when a
+    graph is explicitly streamed with `stream_mode="messages"` inside a node."""
+
+    def respond(state: MessagesState) -> dict:
+        return {"messages": [AIMessage("hi there", id="ai1")]}
+
+    subgraph = (
+        StateGraph(MessagesState)
+        .add_node("respond", respond)
+        .add_edge(START, "respond")
+        .compile()
+    )
+    msgs = {"messages": [("user", "hello")]}
+
+    top_level = [msg for msg, _ in subgraph.stream(msgs, stream_mode="messages")]
+
+    def parent_node(state: MessagesState, config: RunnableConfig) -> dict:
+        nonlocal in_node
+        in_node = [
+            msg for msg, _ in subgraph.stream(msgs, config, stream_mode="messages")
+        ]
+        return {}
+
+    in_node: list = []
+    StateGraph(MessagesState).add_node("parent", parent_node).add_edge(
+        START, "parent"
+    ).compile().invoke(msgs)
+
+    assert [m.content for m in top_level] == ["hi there"]
+    assert [m.content for m in in_node] == ["hi there"]
+
+
 def test_get_graph_nonterminal_last_step_source(snapshot: SnapshotAssertion) -> None:
    class State(TypedDict):
        messages: list[str]
@@ -5809,6 +5809,60 @@ async def test_store_injected_async(
    )  # still overwriting the same one


+@pytest.mark.parametrize("stream_mode", ["tasks", "debug"])
+async def test_failing_task_result_is_streamed_before_the_error(
+    stream_mode: str,
+) -> None:
+    class State(TypedDict):
+        x: str
+
+    async def failing(state: State) -> State:
+        raise ValueError("boom")
+
+    graph = StateGraph(State).add_node("failing", failing).add_edge(START, "failing")
+    errors = []
+    with pytest.raises(ValueError, match="boom"):
+        async for chunk in graph.compile().astream({"x": ""}, stream_mode=stream_mode):
+            payload = chunk["payload"] if stream_mode == "debug" else chunk
+            if "error" in payload:
+                errors.append(payload["error"])
+
+    assert [type(error) for error in errors] == [ValueError]
+
+
+@pytest.mark.parametrize(
+    ("parallel", "stream_mode"), [(False, "custom"), (True, "values")]
+)
+async def test_graph_error_handler_recovers_on_the_executor_path(
+    parallel: bool, stream_mode: str
+) -> None:
+    class State(TypedDict):
+        x: str
+
+    async def failing(state: State) -> State:
+        raise ValueError("boom")
+
+    async def handler(state: State) -> Command:
+        return Command(update={"x": "handled"}, goto=END)
+
+    builder = (
+        StateGraph(State)
+        .add_node("failing", failing, error_handler=handler)
+        .add_edge(START, "failing")
+    )
+    if parallel:
+        builder.add_node("sibling", lambda state: {}).add_edge(START, "sibling")
+
+    chunks = [
+        chunk
+        async for chunk in builder.compile().astream(
+            {"x": ""}, stream_mode=[stream_mode, "values"]
+        )
+    ]
+
+    assert [c[-1] for c in chunks if c[0] == "values"][-1] == {"x": "handled"}
+
+
 async def test_debug_retry(async_checkpointer: BaseCheckpointSaver):
    class State(TypedDict):
        messages: Annotated[list[str], operator.add]
@@ -5875,6 +5929,55 @@ async def test_debug_retry(async_checkpointer: BaseCheckpointSaver):
        assert stream_parent_conf == history_parent_conf


+async def test_interrupt_after_streams_the_checkpoint_it_pauses_on(
+    async_checkpointer: BaseCheckpointSaver,
+) -> None:
+    class State(TypedDict):
+        x: int
+
+    graph = (
+        StateGraph(State)
+        .add_node("one", lambda state: {"x": 1})
+        .add_node("two", lambda state: {"x": 2})
+        .add_edge(START, "one")
+        .add_edge("one", "two")
+        .compile(checkpointer=async_checkpointer, interrupt_after=["one"])
+    )
+    config = {"configurable": {"thread_id": "1"}}
+
+    checkpoints = [
+        chunk
+        async for chunk in graph.astream({"x": 0}, config, stream_mode="checkpoints")
+    ]
+
+    state = await graph.aget_state(config)
+    assert checkpoints[-1]["config"] == state.config
+    assert checkpoints[-1]["next"] == ["two"]
+
+
+async def test_recursion_limit_streams_the_last_checkpoint_saved(
+    async_checkpointer: BaseCheckpointSaver,
+) -> None:
+    class State(TypedDict):
+        x: int
+
+    graph = (
+        StateGraph(State)
+        .add_node("loop", lambda state: {"x": state["x"] + 1})
+        .add_edge(START, "loop")
+        .add_edge("loop", "loop")
+        .compile(checkpointer=async_checkpointer)
+    )
+    config = {"configurable": {"thread_id": "1"}, "recursion_limit": 3}
+    checkpoints = []
+
+    with pytest.raises(GraphRecursionError):
+        async for chunk in graph.astream({"x": 0}, config, stream_mode="checkpoints"):
+            checkpoints.append(chunk)
+
+    assert checkpoints[-1]["config"] == (await graph.aget_state(config)).config
+
+
 async def test_debug_subgraphs(
    async_checkpointer: BaseCheckpointSaver, durability: Durability
 ):
@@ -6731,6 +6834,31 @@ async def test_multiple_updates() -> None:
    ]


+@NEEDS_CONTEXTVARS
+async def test_task_in_subgraph_node_returns_its_value_with_tools_mode() -> None:
+    class State(TypedDict):
+        out: int | None
+
+    @task
+    async def double(x: int) -> int:
+        return x * 2
+
+    async def node(state: State) -> State:
+        return {"out": await double(21)}
+
+    inner = StateGraph(State).add_node("node", node).add_edge(START, "node").compile()
+    graph = (
+        StateGraph(State).add_node("inner", inner).add_edge(START, "inner").compile()
+    )
+
+    chunks = [
+        chunk
+        async for chunk in graph.astream({"out": None}, stream_mode=["values", "tools"])
+    ]
+
+    assert chunks[-1] == ("values", {"out": 42})
+
+
@NEEDS_CONTEXTVARS
 async def test_falsy_return_from_task(async_checkpointer: BaseCheckpointSaver) -> None:
    """Test with a falsy return from a task."""
@@ -7495,6 +7623,40 @@ async def test_entrypoint_from_async_generator() -> None:
            yield "b"


+@NEEDS_CONTEXTVARS
+@pytest.mark.parametrize("timeout", [None, 30])
+@pytest.mark.parametrize("version", ["v1", "v2"])
+async def test_awaited_task_returns_its_value_under_astream_events(
+    version: Literal["v1", "v2"], timeout: int | None
+) -> None:
+    class State(TypedDict):
+        doubled: int
+        zeroed: int
+
+    @task(timeout=timeout)
+    async def double(x: int) -> int:
+        return x * 2
+
+    @task(timeout=timeout)
+    async def zero() -> int:
+        return 0
+
+    async def node(state: State) -> State:
+        return {"doubled": await double(21), "zeroed": await zero()}
+
+    graph = StateGraph(State).add_node("node", node).add_edge(START, "node").compile()
+
+    outputs = [
+        event["data"]["output"]
+        async for event in graph.astream_events(
+            {"doubled": 0, "zeroed": 1}, version=version
+        )
+        if event["event"] == "on_chain_end" and event["name"] == "node"
+    ]
+
+    assert outputs == [{"doubled": 42, "zeroed": 0}]
+
+
@NEEDS_CONTEXTVARS
 async def test_named_tasks_functional() -> None:
    class Foo:
@@ -9341,6 +9503,44 @@ async def test_subgraph_streaming_async() -> None:
    assert result["num_chunks"] == 9


+async def test_subgraph_streaming_node_output_messages_async() -> None:
+    """Messages returned by a node (not streamed by an LLM) are emitted when a
+    graph is explicitly streamed with `stream_mode="messages"` inside a node."""
+
+    async def respond(state: MessagesState) -> dict:
+        return {"messages": [AIMessage("hi there", id="ai1")]}
+
+    subgraph = (
+        StateGraph(MessagesState)
+        .add_node("respond", respond)
+        .add_edge(START, "respond")
+        .compile()
+    )
+    msgs = {"messages": [("user", "hello")]}
+
+    top_level = [msg async for msg, _ in subgraph.astream(msgs, stream_mode="messages")]
+
+    async def parent_node(state: MessagesState, config: RunnableConfig) -> dict:
+        nonlocal in_node
+        in_node = [
+            msg
+            async for msg, _ in subgraph.astream(msgs, config, stream_mode="messages")
+        ]
+        return {}
+
+    in_node: list = []
+    await (
+        StateGraph(MessagesState)
+        .add_node("parent", parent_node)
+        .add_edge(START, "parent")
+        .compile()
+        .ainvoke(msgs)
+    )
+
+    assert [m.content for m in top_level] == ["hi there"]
+    assert [m.content for m in in_node] == ["hi there"]
+
+
@NEEDS_CONTEXTVARS
 async def test_null_resume_disallowed_with_multiple_interrupts(
    async_checkpointer: BaseCheckpointSaver,
@@ -16,9 +16,11 @@ from langgraph_sdk.schema import StreamPart
 from pydantic import BaseModel
 from typing_extensions import TypedDict

+from langgraph._internal._constants import CONFIG_KEY_STREAM
 from langgraph.errors import GraphInterrupt
 from langgraph.graph import END, START, MessagesState, StateGraph, add_messages
 from langgraph.pregel import Pregel
+from langgraph.pregel.protocol import StreamProtocol
 from langgraph.pregel.remote import RemoteGraph
 from langgraph.types import Interrupt, StateSnapshot
 from tests.any_str import AnyStr
@@ -819,6 +821,131 @@ async def test_astream():
    ]


+_CALLER_CONFIG: RunnableConfig = {
+    "configurable": {"thread_id": "thread_1", "checkpoint_ns": "outer:123"}
+}
+_SUBGRAPH_PARTS = [
+    StreamPart(event="values|sub:abc", data={"chunk": "inner"}),
+    StreamPart(event="values", data={"chunk": "root"}),
+]
+_CALLER_PREFIXED = [
+    (("outer:123", "sub:abc"), {"chunk": "inner"}),
+    (("outer:123",), {"chunk": "root"}),
+]
+
+
+def test_stream_subgraphs_keeps_caller_namespace():
+    mock_sync_client = MagicMock()
+    mock_sync_client.runs.stream.return_value = _SUBGRAPH_PARTS
+    remote_pregel = RemoteGraph("test_graph_id", sync_client=mock_sync_client)
+
+    stream_parts = list(
+        remote_pregel.stream(
+            {"input": "data"}, _CALLER_CONFIG, stream_mode="values", subgraphs=True
+        )
+    )
+
+    assert stream_parts == _CALLER_PREFIXED
+
+
+async def test_astream_subgraphs_keeps_caller_namespace():
+    mock_async_client = MagicMock()
+    async_iter = MagicMock()
+    async_iter.__aiter__.return_value = _SUBGRAPH_PARTS
+    mock_async_client.runs.stream.return_value = async_iter
+    remote_pregel = RemoteGraph("test_graph_id", client=mock_async_client)
+
+    stream_parts = [
+        part
+        async for part in remote_pregel.astream(
+            {"input": "data"}, _CALLER_CONFIG, stream_mode="values", subgraphs=True
+        )
+    ]
+
+    assert stream_parts == _CALLER_PREFIXED
+
+
+# Subgraph events requested only on behalf of a streaming parent graph
+_PARENT_STREAMED_PARTS = [
+    StreamPart(event="updates|sub:abc", data={"inner": {"chunk": "inner"}}),
+    StreamPart(event="messages|sub:abc", data=[{"content": "token"}, {}]),
+    StreamPart(event="updates", data={"sub": {"chunk": "root"}}),
+]
+
+
+def _parent_streamed_config(parent_parts: list) -> RunnableConfig:
+    return {
+        "configurable": {
+            "thread_id": "thread_1",
+            "checkpoint_ns": "outer:123",
+            CONFIG_KEY_STREAM: StreamProtocol(
+                parent_parts.append, {"updates", "messages"}
+            ),
+        }
+    }
+
+
+def test_stream_without_subgraphs_skips_parent_requested_subgraph_events():
+    mock_sync_client = MagicMock()
+    mock_sync_client.runs.stream.return_value = _PARENT_STREAMED_PARTS
+    remote_pregel = RemoteGraph("test_graph_id", sync_client=mock_sync_client)
+
+    for stream_mode, expected in [
+        ("updates", [{"sub": {"chunk": "root"}}]),
+        # messages are not filtered: the server dedupes them across namespaces
+        ("messages", [({"content": "token"}, {})]),
+    ]:
+        parent_parts: list = []
+        stream_parts = list(
+            remote_pregel.stream(
+                {"input": "data"},
+                _parent_streamed_config(parent_parts),
+                stream_mode=stream_mode,
+            )
+        )
+
+        assert stream_parts == expected
+        # subgraph events are still requested and forwarded to the parent
+        assert mock_sync_client.runs.stream.call_args.kwargs["stream_subgraphs"]
+        assert [ns for ns, _, _ in parent_parts] == [
+            ("outer:123", "sub:abc"),
+            ("outer:123", "sub:abc"),
+            ("outer:123",),
+        ]
+
+
+async def test_astream_without_subgraphs_skips_parent_requested_subgraph_events():
+    mock_async_client = MagicMock()
+    async_iter = MagicMock()
+    async_iter.__aiter__.return_value = _PARENT_STREAMED_PARTS
+    mock_async_client.runs.stream.return_value = async_iter
+    remote_pregel = RemoteGraph("test_graph_id", client=mock_async_client)
+
+    for stream_mode, expected in [
+        ("updates", [{"sub": {"chunk": "root"}}]),
+        # messages are not filtered: the server dedupes them across namespaces
+        ("messages", [({"content": "token"}, {})]),
+    ]:
+        parent_parts: list = []
+        stream_parts = [
+            part
+            async for part in remote_pregel.astream(
+                {"input": "data"},
+                _parent_streamed_config(parent_parts),
+                stream_mode=stream_mode,
+            )
+        ]
+
+        assert stream_parts == expected
+        # subgraph events are still requested and forwarded to the parent
+        assert mock_async_client.runs.stream.call_args.kwargs["stream_subgraphs"]
+        assert [ns for ns, _, _ in parent_parts] == [
+            ("outer:123", "sub:abc"),
+            ("outer:123", "sub:abc"),
+            ("outer:123",),
+        ]
+
+
 def test_invoke():
    # set up test
    mock_sync_client = MagicMock()
@@ -2097,6 +2097,47 @@ def test_graph_error_handler_can_route_with_command():
    assert result["foo"] == "handled_next"


+# Each case runs the failing node on the executor path instead of the
+# single-task fast path.
+@pytest.mark.parametrize(
+    ("parallel", "stream_mode", "subgraphs"),
+    [
+        (False, "custom", False),
+        (False, "messages", False),
+        (False, "values", True),
+        (True, "values", False),
+    ],
+)
+def test_graph_error_handler_recovers_on_the_executor_path(
+    parallel: bool, stream_mode: str, subgraphs: bool
+) -> None:
+    class State(TypedDict):
+        x: str
+
+    def failing(state: State) -> State:
+        raise ValueError("boom")
+
+    def handler(state: State) -> Command:
+        return Command(update={"x": "handled"}, goto=END)
+
+    builder = (
+        StateGraph(State)
+        .add_node("failing", failing, error_handler=handler)
+        .add_edge(START, "failing")
+    )
+    if parallel:
+        builder.add_node("sibling", lambda state: {}).add_edge(START, "sibling")
+    graph = builder.compile()
+
+    chunks = list(
+        graph.stream(
+            {"x": ""}, stream_mode=[stream_mode, "values"], subgraphs=subgraphs
+        )
+    )
+
+    assert [c[-1] for c in chunks if c[-2] == "values"][-1] == {"x": "handled"}
+
+
 def test_graph_error_handler_failure_fails_run():
    class State(TypedDict):
        foo: str
@@ -0,0 +1,339 @@
+"""Tests for `Runtime.on_exit`: cleanup that runs however a graph run ends."""
+
+from __future__ import annotations
+
+import asyncio
+import sys
+
+import pytest
+from langgraph.checkpoint.memory import InMemorySaver
+from typing_extensions import TypedDict
+
+from langgraph.graph import END, START, StateGraph
+from langgraph.runtime import Runtime, get_runtime
+from langgraph.types import Command, interrupt
+
+pytestmark = pytest.mark.anyio
+
+NEEDS_CONTEXTVARS = pytest.mark.skipif(
+    sys.version_info < (3, 11),
+    reason="Python 3.11+ is required for async contextvars support",
+)
+
+
+class State(TypedDict, total=False):
+    value: str
+
+
+def _two_node_graph(first, second=None, *, checkpointer=None):
+    builder = StateGraph(State)
+    builder.add_node("first", first)
+    builder.add_edge(START, "first")
+    if second is None:
+        builder.add_edge("first", END)
+    else:
+        builder.add_node("second", second)
+        builder.add_edge("first", "second")
+        builder.add_edge("second", END)
+    return builder.compile(checkpointer=checkpointer)
+
+
+# sync
+
+
+def test_runs_on_success_in_lifo_order() -> None:
+    events: list[str] = []
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup-a"))
+        runtime.on_exit(lambda: events.append("cleanup-b"))
+        events.append("first")
+        return {"value": "x"}
+
+    def second(state: State) -> State:
+        events.append("second")
+        return {}
+
+    graph = _two_node_graph(first, second)
+    assert graph.invoke({}) == {"value": "x"}
+    assert events == ["first", "second", "cleanup-b", "cleanup-a"]
+
+
+def test_runs_on_error_and_keeps_the_run_exception() -> None:
+    events: list[str] = []
+
+    def failing_cleanup() -> None:
+        events.append("failing")
+        raise KeyError("cleanup failed")
+
+    def first(state: State) -> State:
+        runtime = get_runtime()
+        runtime.on_exit(lambda: events.append("cleanup"))
+        runtime.on_exit(failing_cleanup)
+        raise ValueError("node failed")
+
+    graph = _two_node_graph(first)
+    with pytest.raises(ValueError, match="node failed"):
+        graph.invoke({})
+    # the failing callback did not stop the earlier one from running
+    assert events == ["failing", "cleanup"]
+
+
+def test_callback_error_is_raised_when_the_run_succeeded() -> None:
+    def failing_cleanup() -> None:
+        raise KeyError("cleanup failed")
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(failing_cleanup)
+        return {}
+
+    graph = _two_node_graph(first)
+    with pytest.raises(KeyError, match="cleanup failed"):
+        graph.invoke({})
+
+
+def test_runs_on_interrupt() -> None:
+    events: list[str] = []
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup"))
+        return {"value": interrupt("approve?")}
+
+    graph = _two_node_graph(first, checkpointer=InMemorySaver())
+    config = {"configurable": {"thread_id": "1"}}
+    result = graph.invoke({}, config)
+    assert "__interrupt__" in result
+    assert events == ["cleanup"]
+
+    # each resumed run is its own invocation with its own callbacks
+    graph.invoke(Command(resume="ok"), config)
+    assert events == ["cleanup", "cleanup"]
+
+
+def test_runs_when_the_stream_is_closed_early() -> None:
+    events: list[str] = []
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup"))
+        return {"value": "x"}
+
+    def second(state: State) -> State:
+        events.append("second")
+        return {}
+
+    graph = _two_node_graph(first, second)
+    stream = graph.stream({}, stream_mode="updates")
+    assert next(stream) == {"first": {"value": "x"}}
+    assert events == []
+    stream.close()
+    assert events == ["cleanup"]
+
+
+def test_runs_when_breaking_out_of_stream() -> None:
+    events: list[str] = []
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup"))
+        return {"value": "x"}
+
+    graph = _two_node_graph(first, lambda s: {})
+    for _ in graph.stream({}, stream_mode="updates"):
+        break
+    # the dropped generator is closed by refcounting on CPython
+    assert events == ["cleanup"]
+
+
+def test_subgraph_registers_on_the_root_run() -> None:
+    events: list[str] = []
+
+    def inner(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup"))
+        events.append("inner")
+        return {}
+
+    sub = StateGraph(State)
+    sub.add_node("inner", inner)
+    sub.add_edge(START, "inner")
+    subgraph = sub.compile()
+
+    def call_subgraph(state: State) -> State:
+        subgraph.invoke(state)
+        events.append("subgraph returned")
+        return {}
+
+    def after(state: State) -> State:
+        events.append("after")
+        return {}
+
+    builder = StateGraph(State)
+    builder.add_node("as_node", subgraph)
+    builder.add_node("call_subgraph", call_subgraph)
+    builder.add_node("after", after)
+    builder.add_edge(START, "as_node")
+    builder.add_edge("as_node", "call_subgraph")
+    builder.add_edge("call_subgraph", "after")
+    graph = builder.compile()
+
+    graph.invoke({})
+    # nothing runs when either subgraph ends; both run when the root ends
+    assert events == [
+        "inner",
+        "inner",
+        "subgraph returned",
+        "after",
+        "cleanup",
+        "cleanup",
+    ]
+
+
+def test_async_callback_rejected_in_sync_run() -> None:
+    async def acleanup() -> None:
+        pass
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(acleanup)
+        return {}
+
+    graph = _two_node_graph(first)
+    with pytest.raises(TypeError, match="async callback in a synchronous"):
+        graph.invoke({})
+
+
+def test_outside_a_run_raises() -> None:
+    with pytest.raises(RuntimeError, match="during a graph run"):
+        Runtime().on_exit(lambda: None)
+
+
+def test_each_invocation_gets_its_own_callbacks() -> None:
+    calls: list[int] = []
+
+    def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: calls.append(1))
+        return {}
+
+    graph = _two_node_graph(first)
+    graph.invoke({})
+    graph.invoke({})
+    assert calls == [1, 1]
+
+
+# async
+
+
+async def test_async_runs_on_success_sync_and_async_callbacks() -> None:
+    events: list[str] = []
+
+    async def acleanup() -> None:
+        await asyncio.sleep(0)
+        events.append("async-cleanup")
+
+    async def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("sync-cleanup"))
+        runtime.on_exit(acleanup)
+        events.append("first")
+        return {"value": "x"}
+
+    graph = _two_node_graph(first)
+    assert await graph.ainvoke({}) == {"value": "x"}
+    assert events == ["first", "async-cleanup", "sync-cleanup"]
+
+
+async def test_async_runs_on_error_and_keeps_the_run_exception() -> None:
+    events: list[str] = []
+
+    async def failing_cleanup() -> None:
+        raise KeyError("cleanup failed")
+
+    async def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup"))
+        runtime.on_exit(failing_cleanup)
+        raise ValueError("node failed")
+
+    graph = _two_node_graph(first)
+    with pytest.raises(ValueError, match="node failed"):
+        await graph.ainvoke({})
+    assert events == ["cleanup"]
+
+
+async def test_async_runs_on_interrupt() -> None:
+    events: list[str] = []
+
+    async def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(lambda: events.append("cleanup"))
+        return {"value": interrupt("approve?")}
+
+    graph = _two_node_graph(first, checkpointer=InMemorySaver())
+    result = await graph.ainvoke({}, {"configurable": {"thread_id": "1"}})
+    assert "__interrupt__" in result
+    assert events == ["cleanup"]
+
+
+async def test_async_runs_on_cancel_after_the_node_unwinds() -> None:
+    events: list[str] = []
+    started = asyncio.Event()
+
+    async def acleanup() -> None:
+        events.append("cleanup")
+
+    async def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(acleanup)
+        started.set()
+        try:
+            await asyncio.sleep(60)
+        finally:
+            events.append("node finally")
+        return {}
+
+    graph = _two_node_graph(first)
+    task = asyncio.create_task(graph.ainvoke({}))
+    await started.wait()
+    task.cancel()
+    with pytest.raises(asyncio.CancelledError):
+        await task
+    # cleanup ran after the node had stopped
+    assert events == ["node finally", "cleanup"]
+
+
+async def test_async_runs_when_the_stream_is_closed_early() -> None:
+    events: list[str] = []
+
+    async def acleanup() -> None:
+        events.append("cleanup")
+
+    async def first(state: State, runtime: Runtime) -> State:
+        runtime.on_exit(acleanup)
+        return {"value": "x"}
+
+    async def second(state: State) -> State:
+        events.append("second")
+        return {}
+
+    graph = _two_node_graph(first, second)
+    stream = graph.astream({}, stream_mode="updates")
+    assert await stream.__anext__() == {"first": {"value": "x"}}
+    assert events == []
+    await stream.aclose()
+    assert events == ["cleanup"]
+
+
+@NEEDS_CONTEXTVARS
+async def test_async_subgraph_registers_on_the_root_run() -> None:
+    events: list[str] = []
+
+    async def inner(state: State) -> State:
+        get_runtime().on_exit(lambda: events.append("cleanup"))
+        return {}
+
+    sub = StateGraph(State)
+    sub.add_node("inner", inner)
+    sub.add_edge(START, "inner")
+    subgraph = sub.compile()
+
+    async def call_subgraph(state: State) -> State:
+        await subgraph.ainvoke(state)
+        events.append("subgraph returned")
+        return {}
+
+    graph = _two_node_graph(call_subgraph, lambda s: events.append("after") or {})
+    await graph.ainvoke({})
+    assert events == ["subgraph returned", "after", "cleanup"]
@@ -0,0 +1,308 @@
+import sys
+from typing import Any
+
+import pytest
+from langchain_core.language_models.fake_chat_models import GenericFakeChatModel
+from langchain_core.messages import AIMessage
+from langchain_core.tools import tool
+
+from langgraph.func import entrypoint, task
+from langgraph.graph import START, MessagesState, StateGraph
+from langgraph.graph.message import push_message
+from langgraph.graph.state import CompiledStateGraph
+
+pytestmark = pytest.mark.anyio
+
+NEEDS_CONTEXTVARS = pytest.mark.skipif(
+    sys.version_info < (3, 11),
+    reason="Python 3.11+ is required for async contextvars support",
+)
+
+
+@tool
+def echo(text: str) -> str:
+    """Echo the text."""
+    return text
+
+
+def _one_node_graph(node: Any) -> CompiledStateGraph:
+    return (
+        StateGraph(MessagesState)
+        .add_node(node)
+        .add_edge(START, node.__name__)
+        .compile()
+    )
+
+
+def _chat_and_echo(state: MessagesState) -> dict:
+    echo.invoke({"text": "done"})
+    return {"messages": [GenericFakeChatModel(messages=iter(["one two"])).invoke("hi")]}
+
+
+async def _achat_and_echo(state: MessagesState) -> dict:
+    await echo.ainvoke({"text": "done"})
+    model = GenericFakeChatModel(messages=iter(["one two"]))
+    return {"messages": [await model.ainvoke("hi")]}
+
+
+def _segment_names(ns: tuple[str, ...]) -> tuple[str, ...]:
+    return tuple(segment.split(":")[0] for segment in ns)
+
+
+def _summary(part: dict) -> tuple[tuple[str, ...], str]:
+    if part["type"] == "messages":
+        return _segment_names(part["ns"]), part["data"][0].content
+    return _segment_names(part["ns"]), part["data"]["event"]
+
+
+_CHAT_AND_ECHO_EVENTS = [
+    ((), "tool-started"),
+    ((), "tool-finished"),
+    ((), "one"),
+    ((), " "),
+    ((), "two"),
+]
+
+_CHILD_CONFIGS = pytest.mark.parametrize(
+    "child_config",
+    [
+        None,
+        {"configurable": {"thread_id": "child"}},
+        {"configurable": {"thread_id": "parent"}},
+    ],
+    ids=["no_config", "own_thread", "caller_thread"],
+)
+
+
+_TASK_CALLERS = pytest.mark.parametrize(
+    "nested", [False, True], ids=["from_entrypoint", "from_task_in_task"]
+)
+
+
+@_TASK_CALLERS
+@pytest.mark.parametrize("subgraphs", [False, True])
+def test_task_events_stream_under_the_graph_namespace(
+    nested: bool, subgraphs: bool
+) -> None:
+    @task
+    def chat_and_echo() -> None:
+        _chat_and_echo({"messages": []})
+
+    @task
+    def outer() -> None:
+        chat_and_echo().result()
+
+    @entrypoint()
+    def workflow(inputs: dict) -> None:
+        (outer() if nested else chat_and_echo()).result()
+
+    parts = workflow.stream(
+        {}, stream_mode=["messages", "tools"], subgraphs=subgraphs, version="v2"
+    )
+
+    assert [_summary(part) for part in parts] == _CHAT_AND_ECHO_EVENTS
+
+
+@NEEDS_CONTEXTVARS
+@_TASK_CALLERS
+@pytest.mark.parametrize("subgraphs", [False, True])
+async def test_task_events_stream_under_the_graph_namespace_async(
+    nested: bool, subgraphs: bool
+) -> None:
+    @task
+    async def chat_and_echo() -> None:
+        await _achat_and_echo({"messages": []})
+
+    @task
+    async def outer() -> None:
+        await chat_and_echo()
+
+    @entrypoint()
+    async def workflow(inputs: dict) -> None:
+        await (outer() if nested else chat_and_echo())
+
+    parts = workflow.astream(
+        {}, stream_mode=["messages", "tools"], subgraphs=subgraphs, version="v2"
+    )
+
+    assert [_summary(part) async for part in parts] == _CHAT_AND_ECHO_EVENTS
+
+
+def test_task_events_stream_under_the_namespace_of_their_subgraph() -> None:
+    @task
+    def chat_and_echo() -> None:
+        _chat_and_echo({"messages": []})
+
+    def node(state: MessagesState) -> dict:
+        chat_and_echo().result()
+        return {}
+
+    sub = _one_node_graph(node)
+    parent = StateGraph(MessagesState).add_node("sub", sub).add_edge(START, "sub")
+
+    parts = parent.compile().stream(
+        {"messages": []},
+        stream_mode=["messages", "tools"],
+        subgraphs=True,
+        version="v2",
+    )
+
+    assert [_summary(part) for part in parts] == [
+        (("sub",), event) for _, event in _CHAT_AND_ECHO_EVENTS
+    ]
+
+
+@_CHILD_CONFIGS
+def test_child_graph_streams_only_to_the_node_streaming_it(
+    child_config: dict | None,
+) -> None:
+    child = _one_node_graph(_chat_and_echo)
+    received = []
+
+    def node(state: MessagesState) -> dict:
+        received.extend(
+            child.stream(
+                {"messages": []}, child_config, stream_mode=["messages", "tools"]
+            )
+        )
+        return {}
+
+    parts = _one_node_graph(node).stream(
+        {"messages": []},
+        {"configurable": {"thread_id": "parent"}},
+        stream_mode=["messages", "tools"],
+    )
+
+    assert list(parts) == []
+    assert len(received) == len(_CHAT_AND_ECHO_EVENTS)
+
+
+@NEEDS_CONTEXTVARS
+@_CHILD_CONFIGS
+async def test_child_graph_streams_only_to_the_node_streaming_it_async(
+    child_config: dict | None,
+) -> None:
+    child = _one_node_graph(_achat_and_echo)
+    received = []
+
+    async def node(state: MessagesState) -> dict:
+        stream = child.astream(
+            {"messages": []}, child_config, stream_mode=["messages", "tools"]
+        )
+        received.extend([part async for part in stream])
+        return {}
+
+    parts = _one_node_graph(node).astream(
+        {"messages": []},
+        {"configurable": {"thread_id": "parent"}},
+        stream_mode=["messages", "tools"],
+    )
+
+    assert [part async for part in parts] == []
+    assert len(received) == len(_CHAT_AND_ECHO_EVENTS)
+
+
+@pytest.mark.parametrize(
+    "child_config",
+    [
+        {"configurable": {"thread_id": "child"}},
+        {"configurable": {"thread_id": "parent"}},
+    ],
+    ids=["own_thread", "caller_thread"],
+)
+def test_child_graph_on_its_own_thread_skips_a_caller_streaming_subgraphs(
+    child_config: dict,
+) -> None:
+    child = _one_node_graph(_chat_and_echo)
+
+    def node(state: MessagesState) -> dict:
+        child.invoke({"messages": []}, child_config)
+        return {}
+
+    parts = _one_node_graph(node).stream(
+        {"messages": []},
+        {"configurable": {"thread_id": "parent"}},
+        stream_mode=["messages", "tools"],
+        subgraphs=True,
+    )
+
+    assert list(parts) == []
+
+
+def test_node_gets_the_events_of_every_subgraph_it_streams() -> None:
+    sub = _one_node_graph(_chat_and_echo)
+    counts = []
+
+    def node(state: MessagesState) -> dict:
+        for _ in range(2):
+            stream = sub.stream({"messages": []}, stream_mode=["messages", "tools"])
+            counts.append(len(list(stream)))
+        return {}
+
+    _one_node_graph(node).invoke({"messages": []})
+
+    assert counts == [len(_CHAT_AND_ECHO_EVENTS)] * 2
+
+
+@NEEDS_CONTEXTVARS
+async def test_node_gets_the_events_of_every_subgraph_it_streams_async() -> None:
+    sub = _one_node_graph(_achat_and_echo)
+    counts = []
+
+    async def node(state: MessagesState) -> dict:
+        for _ in range(2):
+            stream = sub.astream({"messages": []}, stream_mode=["messages", "tools"])
+            counts.append(len([part async for part in stream]))
+        return {}
+
+    await _one_node_graph(node).ainvoke({"messages": []})
+
+    assert counts == [len(_CHAT_AND_ECHO_EVENTS)] * 2
+
+
+def _push(state: MessagesState) -> dict:
+    push_message(AIMessage(id="pushed", content="hi"), state_key=None)
+    return {}
+
+
+def _namespaces(graph: CompiledStateGraph, subgraphs: bool) -> list[tuple[str, ...]]:
+    parts = graph.stream(
+        {"messages": []}, stream_mode="messages", subgraphs=subgraphs, version="v2"
+    )
+    return [_segment_names(part["ns"]) for part in parts]
+
+
+def _streams_pushing_graph(received: list) -> Any:
+    pushing = _one_node_graph(_push)
+
+    def node(state: MessagesState) -> dict:
+        for message, _ in pushing.stream({"messages": []}, stream_mode="messages"):
+            received.append(message.id)
+        return {}
+
+    return node
+
+
+def test_pushed_message_streams_under_the_namespace_of_its_graph() -> None:
+    pushing = _one_node_graph(_push)
+    parent = StateGraph(MessagesState).add_node("sub", pushing).add_edge(START, "sub")
+
+    assert _namespaces(pushing, subgraphs=True) == [()]
+    assert _namespaces(parent.compile(), subgraphs=False) == []
+    assert _namespaces(parent.compile(), subgraphs=True) == [("sub",)]
+
+
+def test_pushed_message_streams_to_the_node_that_streams_its_graph() -> None:
+    received: list[str] = []
+    graph = _one_node_graph(_streams_pushing_graph(received))
+
+    assert _namespaces(graph, subgraphs=False) == []
+    assert received == ["pushed"]
+
+
+def test_pushed_message_streams_to_every_stream_that_takes_it() -> None:
+    received: list[str] = []
+    graph = _one_node_graph(_streams_pushing_graph(received))
+
+    assert _namespaces(graph, subgraphs=True) == [("node",)]
+    assert received == ["pushed"]
@@ -13,13 +13,17 @@ from typing import Annotated, Any
 import pytest
 from langchain_core.messages import AIMessage
 from langchain_core.tools import tool
+from langgraph.checkpoint.memory import InMemorySaver
 from langgraph.prebuilt import ToolNode, ToolRuntime
 from typing_extensions import TypedDict

 from langgraph.constants import END, START
+from langgraph.errors import GraphDrained
 from langgraph.graph import StateGraph
 from langgraph.graph.message import add_messages
 from langgraph.pregel._tools import _tool_call_writer
+from langgraph.runtime import RunControl
+from langgraph.types import Command, interrupt


 class _State(TypedDict):
@@ -54,14 +58,14 @@ def _caller_async(tool_name: str, tool_args: dict[str, Any], tc_id: str = "tc1")
    return caller


-def _build_graph(caller, tools) -> Any:
+def _build_graph(caller, tools, checkpointer=None) -> Any:
    sg = StateGraph(_State)
    sg.add_node("caller", caller)
    sg.add_node("tools", ToolNode(tools))
    sg.add_edge(START, "caller")
    sg.add_edge("caller", "tools")
    sg.add_edge("tools", END)
-    return sg.compile()
+    return sg.compile(checkpointer=checkpointer)


 def _tool_events(stream) -> list[tuple[tuple[str, ...], dict]]:
@@ -146,6 +150,105 @@ class TestSyncGraphSyncTool:
        assert kinds == ["tool-started", "tool-error"]
        assert events[1][1]["message"] == "nope"

+    def test_interrupt_in_tool_is_not_a_tool_error(self) -> None:
+        @tool
+        def ask() -> str:
+            """asks."""
+            return interrupt("ok?")
+
+        graph = _build_graph(_caller_sync("ask", {}), [ask], InMemorySaver())
+        config = {"configurable": {"thread_id": "1"}}
+        paused = _tool_events(
+            graph.stream(
+                {"messages": []}, config, stream_mode=["tools"], subgraphs=True
+            )
+        )
+        resumed = _tool_events(
+            graph.stream(
+                Command(resume="yes"), config, stream_mode=["tools"], subgraphs=True
+            )
+        )
+
+        assert [p["event"] for _, p in paused] == ["tool-started"]
+        assert [(p["event"], p["tool_call_id"]) for _, p in resumed] == [
+            ("tool-started", "tc1"),
+            ("tool-finished", "tc1"),
+        ]
+
+    def test_tool_handing_off_to_the_parent_still_ends_its_call(self) -> None:
+        class State(TypedDict):
+            messages: Annotated[list, add_messages]
+            done: bool
+
+        def hand_off(state: State) -> Command:
+            return Command(graph=Command.PARENT, goto="finish", update={"done": True})
+
+        inner = StateGraph(State).add_node("hand_off", hand_off)
+        inner = inner.add_edge(START, "hand_off").compile()
+
+        @tool
+        def handoff() -> str:
+            """hands off."""
+            inner.invoke({"messages": [], "done": False})
+            return "unreachable"
+
+        graph = (
+            StateGraph(State)
+            .add_node("caller", _caller_sync("handoff", {}))
+            .add_node("tools", ToolNode([handoff]))
+            .add_node("finish", lambda state: {"done": True})
+            .add_edge(START, "caller")
+            .add_edge("caller", "tools")
+            .add_edge("tools", END)
+            .add_edge("finish", END)
+            .compile()
+        )
+        events = [
+            payload["event"]
+            for _, payload in _tool_events(
+                graph.stream(
+                    {"messages": [], "done": False},
+                    stream_mode=["tools"],
+                    subgraphs=True,
+                )
+            )
+        ]
+
+        assert events[0] == "tool-started"
+        assert events[-1] == "tool-error", events
+
+    def test_drained_tool_is_not_a_tool_error(self) -> None:
+        control = RunControl()
+
+        def request_drain(state: _State) -> dict:
+            control.request_drain()
+            return {}
+
+        inner = (
+            StateGraph(_State)
+            .add_node("request_drain", request_drain)
+            .add_node("after", lambda state: {})
+            .add_edge(START, "request_drain")
+            .add_edge("request_drain", "after")
+            .compile()
+        )
+
+        @tool
+        def drained() -> str:
+            """drains."""
+            inner.invoke({"messages": []}, control=control)
+            return "unreachable"
+
+        graph = _build_graph(_caller_sync("drained", {}), [drained])
+        chunks: list = []
+        with pytest.raises(GraphDrained):
+            for chunk in graph.stream(
+                {"messages": []}, stream_mode=["tools"], subgraphs=True
+            ):
+                chunks.append(chunk)
+
+        assert [p["event"] for _, p in _tool_events(chunks)] == ["tool-started"]
+
    def test_writer_unset_outside_tool(self) -> None:
        # Outside any tool body the ContextVar that ToolRuntime reads
        # is unset — emitting from there would be a no-op.
@@ -196,6 +299,25 @@ class TestAsyncGraphAsyncTool:
        assert kinds == ["tool-started", "tool-output-delta", "tool-finished"]
        assert events[1][1]["delta"] == "hi"

+    @pytest.mark.anyio
+    async def test_interrupt_in_async_tool_is_not_a_tool_error(self) -> None:
+        @tool
+        async def aask() -> str:
+            """asks."""
+            return interrupt("ok?")
+
+        graph = _build_graph(_caller_async("aask", {}), [aask], InMemorySaver())
+        config = {"configurable": {"thread_id": "1"}}
+        events = []
+        for graph_input in ({"messages": []}, Command(resume="yes")):
+            async for ns, mode, payload in graph.astream(
+                graph_input, config, stream_mode=["tools"], subgraphs=True
+            ):
+                if mode == "tools":
+                    events.append(payload["event"])
+
+        assert events == ["tool-started", "tool-started", "tool-finished"]
+

 class TestConcurrentToolCalls:
    def test_parallel_tool_calls_do_not_bleed(self) -> None:
@@ -1,11 +1,24 @@
 """End-to-end tests for `TracePolicy` input processing on node trace runs."""

+import asyncio
+from collections.abc import AsyncIterator, Iterator
 from typing import Any
+from uuid import UUID

 import pytest
+from langchain_core.callbacks import BaseCallbackHandler
+from langchain_core.language_models.fake_chat_models import FakeMessagesListChatModel
+from langchain_core.messages import AIMessage, HumanMessage
+from langchain_core.runnables import RunnableConfig
 from typing_extensions import TypedDict

-from langgraph.graph import END, START, StateGraph
+from langgraph._internal._runnable import (
+    ASYNCIO_ACCEPTS_CONTEXT,
+    RunnableCallable,
+    RunnableSeq,
+)
+from langgraph.graph import END, START, MessagesState, StateGraph
+from langgraph.graph.state import CompiledStateGraph
 from langgraph.types import TracePolicy
 from tests.fake_tracer import FakeTracer, Run

@@ -147,3 +160,179 @@ async def test_trace_policy_transforms_recorded_inputs_async() -> None:
    run = _node_run(tracer, "n")
    assert run.inputs == {"scrubbed_in": True}
    assert run.outputs == {"value": 6}
+
+
+_RECORDING_POLICY = TracePolicy(
+    process_inputs=lambda _: {"recorded": "inputs"},
+    process_outputs=lambda _: {"recorded": "outputs"},
+)
+
+
+def _chat_graph() -> CompiledStateGraph:
+    model = FakeMessagesListChatModel(
+        responses=[AIMessage("model reply", id="ai-model")]
+    )
+
+    def chat(state: MessagesState) -> dict[str, Any]:
+        reply = model.invoke(state["messages"])
+        return {
+            "messages": [
+                *state["messages"],
+                reply,
+                AIMessage("from node", id="ai-node"),
+            ]
+        }
+
+    return (
+        StateGraph(MessagesState)
+        .add_node("chat", chat, trace_policy=_RECORDING_POLICY)
+        .add_edge(START, "chat")
+        .compile()
+    )
+
+
+def _assert_scrubbed_node_run(tracer: FakeTracer, *, model_is_child: bool) -> None:
+    run = _node_run(tracer, "chat")
+    assert run.inputs == {"recorded": "inputs"}
+    assert run.outputs == {"recorded": "outputs"}
+    if model_is_child:
+        assert [c.name for c in run.child_runs] == ["FakeMessagesListChatModel"]
+
+
+def test_trace_policy_scrubs_tracer_not_messages_stream() -> None:
+    tracer = FakeTracer()
+    streamed = [
+        message.content
+        for message, _ in _chat_graph().stream(
+            {"messages": [HumanMessage("hello", id="h-1")]},
+            {"callbacks": [tracer]},
+            stream_mode="messages",
+        )
+    ]
+
+    assert streamed == ["model reply", "from node"]
+    _assert_scrubbed_node_run(tracer, model_is_child=True)
+
+
+@pytest.mark.anyio
+async def test_trace_policy_scrubs_tracer_not_messages_stream_async() -> None:
+    tracer = FakeTracer()
+    streamed = [
+        message.content
+        async for message, _ in _chat_graph().astream(
+            {"messages": [HumanMessage("hello", id="h-1")]},
+            {"callbacks": [tracer]},
+            stream_mode="messages",
+        )
+    ]
+
+    assert streamed == ["model reply", "from node"]
+    _assert_scrubbed_node_run(tracer, model_is_child=ASYNCIO_ACCEPTS_CONTEXT)
+
+
+@pytest.mark.anyio
+async def test_trace_policy_scrubs_tracer_not_astream_events() -> None:
+    tracer = FakeTracer()
+    chat_events = [
+        event
+        async for event in _chat_graph().astream_events(
+            {"messages": [HumanMessage("hello", id="h-1")]},
+            {"callbacks": [tracer]},
+            version="v2",
+        )
+        if event["name"] == "chat"
+        and event["event"] in ("on_chain_start", "on_chain_end")
+    ]
+
+    assert [e["event"] for e in chat_events] == ["on_chain_start", "on_chain_end"]
+    start, end = (e["data"] for e in chat_events)
+    assert [m.content for m in start["input"]["messages"]] == ["hello"]
+    assert [m.content for m in end["output"]["messages"]] == [
+        "hello",
+        "model reply",
+        "from node",
+    ]
+    _assert_scrubbed_node_run(tracer, model_is_child=ASYNCIO_ACCEPTS_CONTEXT)
+
+
+class _HandlerWithTapMethods(BaseCallbackHandler):
+    def __init__(self) -> None:
+        self.node_payloads: list[Any] = []
+        self.node_run_ids: set[UUID] = set()
+
+    def _record(self, run_id: UUID, payload: Any) -> None:
+        if run_id in self.node_run_ids:
+            self.node_payloads.append(payload)
+
+    def tap_output_iter(self, run_id: UUID, output: Iterator[Any]) -> Iterator[Any]:
+        for chunk in output:
+            self._record(run_id, chunk)
+            yield chunk
+
+    async def tap_output_aiter(
+        self, run_id: UUID, output: AsyncIterator[Any]
+    ) -> AsyncIterator[Any]:
+        async for chunk in output:
+            self._record(run_id, chunk)
+            yield chunk
+
+    def on_chain_start(
+        self, serialized: Any, inputs: Any, *, run_id: UUID, **kwargs: Any
+    ) -> None:
+        if kwargs.get("name") == "n":
+            self.node_run_ids.add(run_id)
+            self._record(run_id, inputs)
+
+    def on_chain_end(self, outputs: Any, *, run_id: UUID, **kwargs: Any) -> None:
+        self._record(run_id, outputs)
+
+
+@pytest.mark.anyio
+@pytest.mark.parametrize("via_events", [False, True], ids=["ainvoke", "astream_events"])
+async def test_trace_policy_reaches_tracers_but_not_custom_handlers(
+    via_events: bool,
+) -> None:
+    graph = (
+        StateGraph(State)
+        .add_node("n", _incr, trace_policy=_RECORDING_POLICY)
+        .add_edge(START, "n")
+        .add_edge("n", END)
+        .compile()
+    )
+    handler = _HandlerWithTapMethods()
+    tracer = FakeTracer()
+    config: RunnableConfig = {"callbacks": [handler, tracer]}
+
+    if via_events:
+        async for _ in graph.astream_events({"value": 1}, config, version="v2"):
+            pass
+    else:
+        assert await graph.ainvoke({"value": 1}, config) == {"value": 2}
+    assert handler.node_payloads == [{"value": 1}, {"value": 2}]
+    run = _node_run(tracer, "n")
+    assert (run.inputs, run.outputs) == (
+        {"recorded": "inputs"},
+        {"recorded": "outputs"},
+    )
+
+
+@pytest.mark.anyio
+async def test_trace_policy_cancel_at_node_end_still_closes_tracer_run() -> None:
+    def cancel_own_task(value: Any) -> Any:
+        asyncio.current_task().cancel()
+        return value
+
+    tracer = FakeTracer()
+    seq = RunnableSeq(
+        RunnableCallable(lambda x: x, trace=False),
+        RunnableCallable(cancel_own_task, trace=False),
+        trace_policy=_RECORDING_POLICY,
+    )
+    with pytest.raises(asyncio.CancelledError):
+        await asyncio.create_task(seq.ainvoke({}, {"callbacks": [tracer]}))
+    for _ in range(100):
+        if tracer.runs:
+            break
+        await asyncio.sleep(0.01)
+
+    assert [r.outputs for r in tracer.runs] == [{"recorded": "outputs"}]
@@ -1,4 +1,6 @@
+import copy
 import functools
+import pickle
 import sys
 import uuid
 from collections.abc import Callable
@@ -25,6 +27,7 @@ from pydantic import BaseModel, Field
 from typing_extensions import NotRequired, Required, TypedDict

 from langgraph._internal._config import (
+    DEFAULT_RECURSION_LIMIT,
    _is_not_empty,
    _merge_callbacks,
    ensure_config,
@@ -620,6 +623,30 @@ def test_merge_configs_merges_metadata_lc_versions() -> None:
    assert merged["metadata"]["lc_agent_name"] == "agent"


+@pytest.mark.parametrize(
+    "rebuild",
+    [
+        lambda config: config,
+        copy.deepcopy,
+        lambda config: pickle.loads(pickle.dumps(config)),
+    ],
+    ids=["as_is", "deepcopy", "pickle"],
+)
+def test_merge_configs_skips_the_injected_default_recursion_limit(
+    rebuild: Callable[[RunnableConfig], RunnableConfig],
+) -> None:
+    merged = merge_configs({"recursion_limit": 50}, rebuild(ensure_config()))
+    assert merged["recursion_limit"] == 50
+
+
+def test_merge_configs_applies_an_explicit_recursion_limit_equal_to_default() -> None:
+    explicit_default = int(DEFAULT_RECURSION_LIMIT)
+    merged = merge_configs(
+        {"recursion_limit": 50}, {"recursion_limit": explicit_default}
+    )
+    assert merged["recursion_limit"] == explicit_default
+
+
 def test_ensure_config_merges_metadata_lc_versions() -> None:
    a = {
        "metadata": {
@@ -772,6 +799,15 @@ def test_ensure_config_tags_concat_preserves_order_and_duplicates() -> None:
    assert merged["tags"] == ["shared", "alpha", "shared", "beta"]


+def test_ensure_config_skips_tags_inherited_from_ambient() -> None:
+    token = var_child_runnable_config.set({"tags": ["inherited"]})
+    try:
+        merged = ensure_config({"tags": ["inherited", "own"]})
+    finally:
+        var_child_runnable_config.reset(token)
+    assert merged["tags"] == ["inherited", "own"]
+
+
 def test_ensure_config_merges_callbacks_across_configs() -> None:
    a_cb = _TrackingCB("a")
    b_cb = _TrackingCB("b")
@@ -1,5 +1,5 @@
 version = 1
-revision = 3
+revision = 5
 requires-python = ">=3.10"
 resolution-markers = [
    "python_full_version >= '3.15'",
@@ -202,11 +202,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.1.4"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e0/2d/a891ca51311197f6ad14a7ef42e2399f36cf2f9bd44752b3dc4eab60fdc5/certifi-2026.1.4.tar.gz", hash = "sha256:ac726dd470482006e014ad384921ed6438c457018f4b3d204aea4281258b2120", size = 154268, upload-time = "2026-01-04T02:42:41.825Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e6/ad/3cc14f097111b4de0040c83a525973216457bbeeb63739ef1ed275c1c021/certifi-2026.1.4-py3-none-any.whl", hash = "sha256:9943707519e4add1115f44c2bc244f782c0249876bf51b6599fee1ffbedd685c", size = 152900, upload-time = "2026-01-04T02:42:40.15Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -931,18 +931,6 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/d2/23/408243171aa9aaba178d3e2559159c24c1171a641aa83b67bdd3394ead8e/idna-3.15-py3-none-any.whl", hash = "sha256:048adeaf8c2d788c40fee287673ccaa74c24ffd8dcf09ffa555a2fbb59f10ac8", size = 72340, upload-time = "2026-05-12T22:45:55.733Z" },
 ]

-[[package]]
-name = "importlib-metadata"
-version = "8.7.1"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "zipp" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/f3/49/3b30cad09e7771a4982d9975a8cbf64f00d4a1ececb53297f1d9a7be1b10/importlib_metadata-8.7.1.tar.gz", hash = "sha256:49fef1ae6440c182052f407c8d34a68f72efc36db9ca90dc0113398f2fdde8bb", size = 57107, upload-time = "2025-12-21T10:00:19.278Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/fa/5e/f8e9a1d23b9c20a551a8a02ea3637b4642e22c2626e3a13a9a29cdea99eb/importlib_metadata-8.7.1-py3-none-any.whl", hash = "sha256:5a1f80bf1daa489495071efbb095d75a634cf28a8bc299581244063b53176151", size = 27865, upload-time = "2025-12-21T10:00:18.329Z" },
-]
-
 [[package]]
 name = "iniconfig"
 version = "2.3.0"
@@ -1451,7 +1439,7 @@ wheels = [

 [[package]]
 name = "langgraph"
-version = "1.2.16"
+version = "1.2.19"
 source = { editable = "." }
 dependencies = [
    { name = "langchain-core" },
@@ -1637,7 +1625,7 @@ wheels = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
    { name = "langchain-core" },
@@ -1734,7 +1722,7 @@ test = [

 [[package]]
 name = "langgraph-checkpoint-sqlite"
-version = "3.1.2"
+version = "3.1.3"
 source = { editable = "../checkpoint-sqlite" }
 dependencies = [
    { name = "aiosqlite" },
@@ -1919,7 +1907,7 @@ requires-dist = [
    { name = "langchain-core", specifier = ">=1.4.0,<2" },
    { name = "langchain-protocol", specifier = ">=0.0.15" },
    { name = "orjson", specifier = ">=3.11.5" },
-    { name = "websockets", specifier = ">=14,<17" },
+    { name = "websockets", specifier = ">=14" },
 ]

 [package.metadata.requires-dev]
@@ -2111,7 +2099,7 @@ wheels = [

 [[package]]
 name = "nbconvert"
-version = "7.17.1"
+version = "7.17.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "beautifulsoup4" },
@@ -2129,9 +2117,9 @@ dependencies = [
    { name = "pygments" },
    { name = "traitlets" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/01/b1/708e53fe2e429c103c6e6e159106bcf0357ac41aa4c28772bd8402339051/nbconvert-7.17.1.tar.gz", hash = "sha256:34d0d0a7e73ce3cbab6c5aae8f4f468797280b01fd8bd2ca746da8569eddd7d2", size = 865311, upload-time = "2026-04-08T00:44:14.914Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/53/6a/252c7620305a9661205aed7a2a4b3592f5207bfca56ad11b7dea2c5ec2f6/nbconvert-7.17.2.tar.gz", hash = "sha256:344a218e8595117ce37396ed025d994f146d076226ebac59255a2fa085ab24d8", size = 865989, upload-time = "2026-10-12T09:11:19.583Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/67/f8/bb0a9d5f46819c821dc1f004aa2cc29b1d91453297dbf5ff20470f00f193/nbconvert-7.17.1-py3-none-any.whl", hash = "sha256:aa85c087b435e7bf1ffd03319f658e285f2b89eccab33bc1ba7025495ab3e7c8", size = 261927, upload-time = "2026-04-08T00:44:12.845Z" },
+    { url = "https://files.pythonhosted.org/packages/81/fd/a8d457af68848a8e5c12dd1d84ff35c415db471af78280e0703db4efbb12/nbconvert-7.17.2-py3-none-any.whl", hash = "sha256:6a143c11c00bf831604da4b1be4347d36ea52367057eb920135d69bc2862fed2", size = 262094, upload-time = "2026-10-12T09:11:17.215Z" },
 ]

 [[package]]
@@ -2189,84 +2177,114 @@ wheels = [

 [[package]]
 name = "opentelemetry-api"
-version = "1.39.1"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "importlib-metadata" },
    { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/97/b9/3161be15bb8e3ad01be8be5a968a9237c3027c5be504362ff800fca3e442/opentelemetry_api-1.39.1.tar.gz", hash = "sha256:fbde8c80e1b937a2c61f20347e91c0c18a1940cecf012d62e65a7caf08967c9c", size = 65767, upload-time = "2025-12-11T13:32:39.182Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/2e/02/6e0ae9cc61bd3169d401077b507b3ebc344745171e1051ab430be012dcd9/opentelemetry_api-1.45.1.tar.gz", hash = "sha256:aa38ed19bcc084ba42782a73255b3582283eced7ad6dddbd6695189e69adfb75", size = 72804, upload-time = "2026-10-06T17:32:58.133Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/cf/df/d3f1ddf4bb4cb50ed9b1139cc7b1c54c34a1e7ce8fd1b9a37c0d1551a6bd/opentelemetry_api-1.39.1-py3-none-any.whl", hash = "sha256:2edd8463432a7f8443edce90972169b195e7d6a05500cd29e6d13898187c9950", size = 66356, upload-time = "2025-12-11T13:32:17.304Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/41/f7dcf80b81ee8e71c1a2b59f14208bc723edbd89ed027a73b175abf6348e/opentelemetry_api-1.45.1-py3-none-any.whl", hash = "sha256:b31553efa588ae44bc306f863c785c5333a9ecc091248c6ee68b4b6c87fdedfb", size = 60256, upload-time = "2026-10-06T17:32:33.506Z" },
+]
+
+[[package]]
+name = "opentelemetry-exporter-http-transport"
+version = "0.66b1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "opentelemetry-api" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/62/0c/e3ebdb4b507f66afcc905e6885a4946969bd75b45988492643356fbbdc63/opentelemetry_exporter_http_transport-0.66b1.tar.gz", hash = "sha256:443080203bf52586ce0b2ad901e8951c61833eab1aa539ae6f1f16fe9e8e7952", size = 11693, upload-time = "2026-10-06T17:32:59.65Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/04/69/6af86ff66492b481c6a4c05dcfd68beb47ed8ba046440a26a2aac76b95c7/opentelemetry_exporter_http_transport-0.66b1-py3-none-any.whl", hash = "sha256:2f95404bdee7f9d2d529c7de56c7bd86d014d774d8fbf137810e0167f8a492bf", size = 12155, upload-time = "2026-10-06T17:32:35.454Z" },
+]
+
+[package.optional-dependencies]
+requests = [
+    { name = "requests" },
+]
+
+[[package]]
+name = "opentelemetry-exporter-otlp-common"
+version = "0.66b1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "opentelemetry-sdk" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/cb/19/41de712173f43057e4532d42ece7d0c6d4210d353e5752433cb14987643f/opentelemetry_exporter_otlp_common-0.66b1.tar.gz", hash = "sha256:6b1403487a2185ac1feb45fd5546fdf8630ce71c36bcefaadf51e2130e9e23f9", size = 14325, upload-time = "2026-10-06T17:33:01.725Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fc/39/8c23d67665c762aa51840fa06f86e902e8f6f1693bc8d7e3d98cd6e2f753/opentelemetry_exporter_otlp_common-0.66b1-py3-none-any.whl", hash = "sha256:00ff8592c3a7cb729ff3fdc7ffa12372c243bdf2163e80c180994d0c7bd83ee9", size = 12385, upload-time = "2026-10-06T17:32:38.177Z" },
 ]

 [[package]]
 name = "opentelemetry-exporter-otlp-proto-common"
-version = "1.39.1"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "opentelemetry-proto" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/e9/9d/22d241b66f7bbde88a3bfa6847a351d2c46b84de23e71222c6aae25c7050/opentelemetry_exporter_otlp_proto_common-1.39.1.tar.gz", hash = "sha256:763370d4737a59741c89a67b50f9e39271639ee4afc999dadfe768541c027464", size = 20409, upload-time = "2025-12-11T13:32:40.885Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/c1/8e/65e85e5137991a3c493b11682151d198638a5bc1dd4b4c5f67e013c57d7c/opentelemetry_exporter_otlp_proto_common-1.45.1.tar.gz", hash = "sha256:2e4adcc3a67bcf57804fc49514f0ef64974ca7590aa3491da389852b4a0628f6", size = 18873, upload-time = "2026-10-06T17:33:04.471Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/8c/02/ffc3e143d89a27ac21fd557365b98bd0653b98de8a101151d5805b5d4c33/opentelemetry_exporter_otlp_proto_common-1.39.1-py3-none-any.whl", hash = "sha256:08f8a5862d64cc3435105686d0216c1365dc5701f86844a8cd56597d0c764fde", size = 18366, upload-time = "2025-12-11T13:32:20.2Z" },
+    { url = "https://files.pythonhosted.org/packages/84/aa/92f225d353904e7f70b8b3e3c1b02db0cf56f744c2e83c581dc372e78873/opentelemetry_exporter_otlp_proto_common-1.45.1-py3-none-any.whl", hash = "sha256:2f446183ae7047b036226f1d846c41a834b0e8755ad13b51a51dd38952eb466c", size = 15393, upload-time = "2026-10-06T17:32:41.911Z" },
 ]

 [[package]]
 name = "opentelemetry-exporter-otlp-proto-http"
-version = "1.39.1"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "googleapis-common-protos" },
    { name = "opentelemetry-api" },
+    { name = "opentelemetry-exporter-http-transport", extra = ["requests"] },
+    { name = "opentelemetry-exporter-otlp-common" },
    { name = "opentelemetry-exporter-otlp-proto-common" },
    { name = "opentelemetry-proto" },
    { name = "opentelemetry-sdk" },
    { name = "requests" },
    { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/80/04/2a08fa9c0214ae38880df01e8bfae12b067ec0793446578575e5080d6545/opentelemetry_exporter_otlp_proto_http-1.39.1.tar.gz", hash = "sha256:31bdab9745c709ce90a49a0624c2bd445d31a28ba34275951a6a362d16a0b9cb", size = 17288, upload-time = "2025-12-11T13:32:42.029Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/1b/17/26487707ea4caa97b17e6e4b5fa72133a53512ffa2f5cf7a49ef284b29cb/opentelemetry_exporter_otlp_proto_http-1.45.1.tar.gz", hash = "sha256:45c218405ce3fd879596924b1874bf9a8f6880206d61065c5a912c8e5c297fb7", size = 28839, upload-time = "2026-10-06T17:33:05.713Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/95/f1/b27d3e2e003cd9a3592c43d099d2ed8d0a947c15281bf8463a256db0b46c/opentelemetry_exporter_otlp_proto_http-1.39.1-py3-none-any.whl", hash = "sha256:d9f5207183dd752a412c4cd564ca8875ececba13be6e9c6c370ffb752fd59985", size = 19641, upload-time = "2025-12-11T13:32:22.248Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/1f/517eaa0187ba106a9da97160ce2add3a371812681dc440930b267f714e42/opentelemetry_exporter_otlp_proto_http-1.45.1-py3-none-any.whl", hash = "sha256:24a97cf3753c7fb52fad44a696e452ff371686339e2acf3309e2eda3d0230700", size = 22180, upload-time = "2026-10-06T17:32:43.946Z" },
 ]

 [[package]]
 name = "opentelemetry-proto"
-version = "1.39.1"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "protobuf" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/49/1d/f25d76d8260c156c40c97c9ed4511ec0f9ce353f8108ca6e7561f82a06b2/opentelemetry_proto-1.39.1.tar.gz", hash = "sha256:6c8e05144fc0d3ed4d22c2289c6b126e03bcd0e6a7da0f16cedd2e1c2772e2c8", size = 46152, upload-time = "2025-12-11T13:32:48.681Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/4b/7f/15f014fb195da6c2dbb6c71399b8e76824878718e94de6454038488eed28/opentelemetry_proto-1.45.1.tar.gz", hash = "sha256:79e0fb95e4616691a469439238aa9224d75779b3e108e895d1aa125ab29ca77c", size = 46488, upload-time = "2026-10-06T17:33:11.49Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/51/95/b40c96a7b5203005a0b03d8ce8cd212ff23f1793d5ba289c87a097571b18/opentelemetry_proto-1.39.1-py3-none-any.whl", hash = "sha256:22cdc78efd3b3765d09e68bfbd010d4fc254c9818afd0b6b423387d9dee46007", size = 72535, upload-time = "2025-12-11T13:32:33.866Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/9a/42ec8180a769516ae757e893b69736826efceac7332553915b4528a91c6d/opentelemetry_proto-1.45.1-py3-none-any.whl", hash = "sha256:f38e2a8413053c180cd3d2637fbb279673ec2f6a6e09c995aafa2f452c52b46e", size = 72488, upload-time = "2026-10-06T17:32:53.057Z" },
 ]

 [[package]]
 name = "opentelemetry-sdk"
-version = "1.39.1"
+version = "1.45.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "opentelemetry-api" },
    { name = "opentelemetry-semantic-conventions" },
    { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/eb/fb/c76080c9ba07e1e8235d24cdcc4d125ef7aa3edf23eb4e497c2e50889adc/opentelemetry_sdk-1.39.1.tar.gz", hash = "sha256:cf4d4563caf7bff906c9f7967e2be22d0d6b349b908be0d90fb21c8e9c995cc6", size = 171460, upload-time = "2025-12-11T13:32:49.369Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a1/79/7392e21a1c8f0c61d90b223e31c7e48cb9d452e91a6b820ad24cca5f23c4/opentelemetry_sdk-1.45.1.tar.gz", hash = "sha256:63d24a6ca645019a631e6a51999c73e93adcac1196ca640b8ae78a7cc4762bf3", size = 218324, upload-time = "2026-10-06T17:33:13.26Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/7c/98/e91cf858f203d86f4eccdf763dcf01cf03f1dae80c3750f7e635bfa206b6/opentelemetry_sdk-1.39.1-py3-none-any.whl", hash = "sha256:4d5482c478513ecb0a5d938dcc61394e647066e0cc2676bee9f3af3f3f45f01c", size = 132565, upload-time = "2025-12-11T13:32:35.069Z" },
+    { url = "https://files.pythonhosted.org/packages/95/3c/87c42b4bd6dd297536f04cd9383d212ac557ecd49f2cbdcd46da1c9ef5c8/opentelemetry_sdk-1.45.1-py3-none-any.whl", hash = "sha256:c604c11dc429810812348989115fa44bd558772a3d7442afc43d024f2c250ca4", size = 140063, upload-time = "2026-10-06T17:32:55.04Z" },
 ]

 [[package]]
 name = "opentelemetry-semantic-conventions"
-version = "0.60b1"
+version = "0.66b1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "opentelemetry-api" },
    { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/91/df/553f93ed38bf22f4b999d9be9c185adb558982214f33eae539d3b5cd0858/opentelemetry_semantic_conventions-0.60b1.tar.gz", hash = "sha256:87c228b5a0669b748c76d76df6c364c369c28f1c465e50f661e39737e84bc953", size = 137935, upload-time = "2025-12-11T13:32:50.487Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/46/e4/dbbfb2a010c4db2224a5114638acede6fe563d33cc20fb1752cebcbe6298/opentelemetry_semantic_conventions-0.66b1.tar.gz", hash = "sha256:497ca63bf383723411e8eaf60c8779e9877633c936bb641080adab59d0eb6ec8", size = 150250, upload-time = "2026-10-06T17:33:14.073Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/7a/5e/5958555e09635d09b75de3c4f8b9cae7335ca545d77392ffe7331534c402/opentelemetry_semantic_conventions-0.60b1-py3-none-any.whl", hash = "sha256:9fa8c8b0c110da289809292b0591220d3a7b53c1526a23021e977d68597893fb", size = 219982, upload-time = "2025-12-11T13:32:36.955Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/14/67f8aa798857f8cf686f515bf93d9bb877ce952ddc8efae0fa25b45ce0d6/opentelemetry_semantic_conventions-0.66b1-py3-none-any.whl", hash = "sha256:d4cddeb4315490b35213f55e2bdc9ac54bb1e4d318927475bed62b35545e581b", size = 206279, upload-time = "2026-10-06T17:32:56.103Z" },
 ]

 [[package]]
@@ -3401,11 +3419,11 @@ wheels = [

 [[package]]
 name = "setuptools"
-version = "83.0.0"
+version = "84.1.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/34/26/f5d29e25ffdb535afef2d35cdb55b325298f96debd670da4c325e08d70f4/setuptools-83.0.0.tar.gz", hash = "sha256:025bccbbf0fa05b6192bc64ae1e7b16e001fd6d6d4d5de03c97b1c1ade523bef", size = 1154254, upload-time = "2026-07-04T15:31:22.699Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/73/eb/7e5f628ee2eb168fecbf8026de7856b2543d6d609cc85f086b9a91ac93d6/setuptools-84.1.0.tar.gz", hash = "sha256:23657205b63d4d31395ab917815247f694aafad28cd21316bf5fd45a51048ea0", size = 1171783, upload-time = "2026-10-19T20:11:43.489Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/5d/40/e1e72872c6354b306daef1703549e8e83b4d43cfea356311bf722a043752/setuptools-83.0.0-py3-none-any.whl", hash = "sha256:29b23c360f22f414dc7336bb39178cc7bcbf6021ed2733cde173f09dba19abb3", size = 1008090, upload-time = "2026-07-04T15:31:20.885Z" },
+    { url = "https://files.pythonhosted.org/packages/18/02/b4bbb90f8f29e334898ff45c6d9b1fffaa6f378017052b3023a06262b5f7/setuptools-84.1.0-py3-none-any.whl", hash = "sha256:5231689c6ac8156cd155cfb98c3fd6fb252c3e7e9bd3d78b64d3c55f132624c9", size = 819380, upload-time = "2026-10-19T20:11:41.684Z" },
 ]

 [[package]]
@@ -4239,15 +4257,6 @@ wheels = [
    { url = "https://files.pythonhosted.org/packages/7d/f3/1ac078fc8fceadcf066469acecacb35d2821cbfaf7d6fc5ac2107c7a314d/xxhash-4.0.1-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:fac4832b638000106207bc44e44b9616a6a416aaee56c62b01d61f3705e49f58", size = 37252, upload-time = "2026-08-17T08:24:06.652Z" },
 ]

-[[package]]
-name = "zipp"
-version = "3.23.0"
-source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e3/02/0f2892c661036d50ede074e376733dca2ae7c6eb617489437771209d4180/zipp-3.23.0.tar.gz", hash = "sha256:a07157588a12518c9d4034df3fbbee09c814741a33ff63c05fa29d26a2404166", size = 25547, upload-time = "2025-06-08T17:06:39.4Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/2e/54/647ade08bf0db230bfea292f893923872fd20be6ac6f53b2b936ba839d75/zipp-3.23.0-py3-none-any.whl", hash = "sha256:071652d6115ed432f5ce1d34c336c0adfd6a884660d1e9712a256d3d3bd4b14e", size = 10276, upload-time = "2025-06-08T17:06:38.034Z" },
-]
-
 [[package]]
 name = "zstandard"
 version = "0.25.0"
@@ -49,11 +49,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.1.4"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/e0/2d/a891ca51311197f6ad14a7ef42e2399f36cf2f9bd44752b3dc4eab60fdc5/certifi-2026.1.4.tar.gz", hash = "sha256:ac726dd470482006e014ad384921ed6438c457018f4b3d204aea4281258b2120", size = 154268, upload-time = "2026-01-04T02:42:41.825Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/e6/ad/3cc14f097111b4de0040c83a525973216457bbeeb63739ef1ed275c1c021/certifi-2026.1.4-py3-none-any.whl", hash = "sha256:9943707519e4add1115f44c2bc244f782c0249876bf51b6599fee1ffbedd685c", size = 152900, upload-time = "2026-01-04T02:42:40.15Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -286,7 +286,7 @@ wheels = [

 [[package]]
 name = "langgraph"
-version = "1.2.16"
+version = "1.2.19"
 source = { editable = "../langgraph" }
 dependencies = [
    { name = "langchain-core" },
@@ -370,7 +370,7 @@ test = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
    { name = "langchain-core" },
@@ -467,7 +467,7 @@ test = [

 [[package]]
 name = "langgraph-checkpoint-sqlite"
-version = "3.1.2"
+version = "3.1.3"
 source = { editable = "../checkpoint-sqlite" }
 dependencies = [
    { name = "aiosqlite" },
@@ -614,7 +614,7 @@ requires-dist = [
    { name = "langchain-core", specifier = ">=1.4.0,<2" },
    { name = "langchain-protocol", specifier = ">=0.0.15" },
    { name = "orjson", specifier = ">=3.11.5" },
-    { name = "websockets", specifier = ">=14,<17" },
+    { name = "websockets", specifier = ">=14" },
 ]

 [package.metadata.requires-dev]
@@ -34,6 +34,14 @@
 - **`thread.output`** — awaitable that resolves to the terminal thread state
  `values` dict after the run lifecycle completes.

+- **`Priority` (beta)** — register an async function with `@priority.assign`
+  and name the object in the `priority` section of `langgraph.json`. The
+  server calls the function with a `PriorityContext` at the first run on a
+  thread with no settings and stores the `PrioritySettings` it returns
+  (`priority`, `fairness_group`, `fairness_weight`) on the thread. Requires
+  `langgraph-api` 0.18.0 or later and a `langgraph-cli` that supports the
+  section; older versions of either ignore it.
+
 ### Changed

 - `client.threads.stream()` now accepts `transport="sse"` (default) or
@@ -2,14 +2,19 @@ from langgraph_sdk.auth import Auth
 from langgraph_sdk.client import get_client, get_sync_client
 from langgraph_sdk.encryption import Encryption
 from langgraph_sdk.encryption.types import DecryptResult, EncryptionContext
+from langgraph_sdk.priority import Priority
+from langgraph_sdk.priority.types import PriorityContext, PrioritySettings

-__version__ = "0.4.8"
+__version__ = "0.4.11"

 __all__ = [
    "Auth",
    "DecryptResult",
    "Encryption",
    "EncryptionContext",
+    "Priority",
+    "PriorityContext",
+    "PrioritySettings",
    "get_client",
    "get_sync_client",
 ]
@@ -14,9 +14,14 @@ import orjson

 from langgraph_sdk._shared.utilities import (
    _orjson_default,
+    _reconnect_delay,
    _validate_reconnect_location,
 )
-from langgraph_sdk.errors import _araise_for_status_typed
+from langgraph_sdk.errors import (
+    InternalServerError,
+    RateLimitError,
+    _araise_for_status_typed,
+)
 from langgraph_sdk.schema import QueryParamTypes, StreamPart
 from langgraph_sdk.sse import SSEDecoder, aiter_lines_raw

@@ -146,7 +151,8 @@ class HttpClient:
        on_response: Callable[[httpx.Response], None] | None = None,
        reconnect_limit: int = 5,
    ) -> Any:
-        """Send a request that automatically reconnects to Location header."""
+        """Send a request that reconnects to its Location header if the response is
+        cut off, with backoff between attempts."""
        request_headers, content = await _aencode_json(json)
        if headers:
            request_headers.update(headers)
@@ -176,13 +182,27 @@ class HttpClient:
                    stacklevel=2,
                )
                await r.aclose()
-                return await self.request_reconnect(
-                    loc,
-                    "GET",
-                    headers=request_headers,
-                    # don't pass on_response so it's only called once
-                    reconnect_limit=reconnect_limit - 1,
-                )
+                for attempt in range(reconnect_limit):
+                    await asyncio.sleep(_reconnect_delay(attempt, base=1.0, cap=5.0))
+                    # A restart or a deploy looks like a refused connection, a
+                    # dropped body, a 429 or a 5xx; other errors won't go away.
+                    try:
+                        return await self.request_reconnect(
+                            loc,
+                            "GET",
+                            headers=headers,
+                            # don't pass on_response so it's only called once
+                            reconnect_limit=0,
+                        )
+                    except httpx.HTTPStatusError as err:
+                        status = err.response.status_code
+                        if attempt == reconnect_limit - 1 or (
+                            status != 429 and status < 500
+                        ):
+                            raise
+                    except httpx.TransportError:
+                        if attempt == reconnect_limit - 1:
+                            raise

    async def stream(
        self,
@@ -218,71 +238,103 @@ class HttpClient:
                request_headers if reconnect_path is None else reconnect_headers
            )
            if last_event_id is not None:
+                current_headers = {
+                    key: value
+                    for key, value in current_headers.items()
+                    if key.lower() != "last-event-id"
+                }
+            if last_event_id:
                current_headers["Last-Event-ID"] = last_event_id
+            if reconnect_path is not None and not any(
+                key.lower() == "last-event-id" and value
+                for key, value in current_headers.items()
+            ):
+                warnings.warn(
+                    "Reconnecting without a Last-Event-ID, so events sent while "
+                    "disconnected are not replayed. Only runs created with "
+                    "stream_resumable=True send the event IDs a reconnect resumes "
+                    "from.",
+                    stacklevel=2,
+                )

            current_method = method if reconnect_path is None else "GET"
            current_content = content if reconnect_path is None else None
            current_params = params if reconnect_path is None else None

-            retry = False
-            async with self.client.stream(
-                current_method,
-                reconnect_path or path,
-                headers=current_headers,
-                content=current_content,
-                params=current_params,
-            ) as res:
-                if reconnect_path is None and on_response:
-                    on_response(res)
-                # check status
-                await _araise_for_status_typed(res)
-                # check content type
-                content_type = res.headers.get("content-type", "").partition(";")[0]
-                if "text/event-stream" not in content_type:
-                    raise httpx.TransportError(
-                        "Expected response header Content-Type to contain 'text/event-stream', "
-                        f"got {content_type!r}"
-                    )
+            reconnect_cause: httpx.HTTPError | None = None
+            try:
+                async with self.client.stream(
+                    current_method,
+                    reconnect_path or path,
+                    headers=current_headers,
+                    content=current_content,
+                    params=current_params,
+                ) as res:
+                    if reconnect_path is None and on_response:
+                        on_response(res)
+                    # check status
+                    await _araise_for_status_typed(res)
+                    # check content type
+                    content_type = res.headers.get("content-type", "").partition(";")[0]
+                    if "text/event-stream" not in content_type:
+                        raise httpx.TransportError(
+                            "Expected response header Content-Type to contain 'text/event-stream', "
+                            f"got {content_type!r}"
+                        )

-                reconnect_location = res.headers.get("location")
-                if reconnect_location:
-                    _validate_reconnect_location(
-                        self.client.base_url, reconnect_location
-                    )
-                    reconnect_path = reconnect_location
+                    reconnect_location = res.headers.get("location")
+                    if reconnect_location:
+                        _validate_reconnect_location(
+                            self.client.base_url, reconnect_location
+                        )
+                        reconnect_path = reconnect_location

-                # parse SSE
-                decoder = SSEDecoder()
-                try:
-                    async for line in aiter_lines_raw(res):
-                        sse = decoder.decode(line=cast("bytes", line).rstrip(b"\n"))
-                        if sse is not None:
+                    # parse SSE
+                    decoder = SSEDecoder(last_event_id=last_event_id)
+                    try:
+                        async for line in aiter_lines_raw(res):
+                            sse = decoder.decode(line=cast("bytes", line).rstrip(b"\n"))
+                            if sse is not None:
+                                if decoder.last_event_id is not None:
+                                    last_event_id = decoder.last_event_id
+                                if sse.event or decoder.has_data:
+                                    # Progress since the last drop earns a fresh
+                                    # set of reconnect attempts.
+                                    reconnect_attempts = 0
+                                    yield sse
+                    except httpx.HTTPError as err:
+                        # httpx.TransportError inherits from HTTPError, so transient
+                        # disconnects during streaming land here.
+                        if reconnect_path is None:
+                            raise
+                        reconnect_cause = err
+                    else:
+                        if sse := decoder.decode(b""):
                            if decoder.last_event_id is not None:
                                last_event_id = decoder.last_event_id
-                            if sse.event or sse.data is not None:
+                            if sse.event or decoder.has_data:
+                                # decoder.decode(b"") flushes the in-flight event and may
+                                # return an empty placeholder when there is no pending
+                                # message. Skip these no-op events so the stream doesn't
+                                # emit a trailing blank item after reconnects.
                                yield sse
-                except httpx.HTTPError:
-                    # httpx.TransportError inherits from HTTPError, so transient
-                    # disconnects during streaming land here.
-                    if reconnect_path is None:
-                        raise
-                    retry = True
-                else:
-                    if sse := decoder.decode(b""):
-                        if decoder.last_event_id is not None:
-                            last_event_id = decoder.last_event_id
-                        if sse.event or sse.data is not None:
-                            # decoder.decode(b"") flushes the in-flight event and may
-                            # return an empty placeholder when there is no pending
-                            # message. Skip these no-op events so the stream doesn't
-                            # emit a trailing blank item after reconnects.
-                            yield sse
-            if retry:
+            except (httpx.TransportError, RateLimitError, InternalServerError) as err:
+                # A reconnect that can't open (connection refused, 429, 5xx or a
+                # non-SSE body) usually means the server is restarting or
+                # mid-deploy, so retry it like a drop. Before the server sends a
+                # Location there is nothing to reconnect to, so raise.
+                if reconnect_path is None:
+                    raise
+                reconnect_cause = err
+            if reconnect_cause is not None:
                reconnect_attempts += 1
                if reconnect_attempts > max_reconnect_attempts:
                    raise httpx.TransportError(
                        "Exceeded maximum SSE reconnection attempts"
-                    )
+                    ) from reconnect_cause
+                await asyncio.sleep(
+                    _reconnect_delay(reconnect_attempts - 1, base=1.0, cap=5.0)
+                )
                continue
            break

@@ -99,6 +99,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v1"] = "v1",
    ) -> AsyncIterator[StreamPart]: ...

@@ -131,6 +132,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v2"],
    ) -> AsyncIterator[StreamPartV2]: ...

@@ -147,6 +149,7 @@ class RunsClient:
        stream_resumable: bool = False,
        metadata: Mapping[str, Any] | None = None,
        config: Config | None = None,
+        context: Context | None = None,
        checkpoint_during: bool | None = None,
        interrupt_before: All | Sequence[str] | None = None,
        interrupt_after: All | Sequence[str] | None = None,
@@ -160,6 +163,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v1"] = "v1",
    ) -> AsyncIterator[StreamPart]: ...

@@ -176,6 +180,7 @@ class RunsClient:
        stream_resumable: bool = False,
        metadata: Mapping[str, Any] | None = None,
        config: Config | None = None,
+        context: Context | None = None,
        checkpoint_during: bool | None = None,
        interrupt_before: All | Sequence[str] | None = None,
        interrupt_after: All | Sequence[str] | None = None,
@@ -189,6 +194,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v2"],
    ) -> AsyncIterator[StreamPartV2]: ...

@@ -385,6 +391,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> Run: ...

    @overload
@@ -414,6 +421,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> Run: ...

    async def create(
@@ -647,6 +655,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> builtins.list[dict] | dict[str, Any]: ...

    @overload
@@ -673,6 +682,7 @@ class RunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> builtins.list[dict] | dict[str, Any]: ...

    async def wait(
@@ -16,7 +16,6 @@ from __future__ import annotations
 import asyncio
 import contextlib
 import logging
-import random
 from collections.abc import AsyncGenerator, AsyncIterator, Generator, Mapping
 from dataclasses import dataclass, field
 from typing import Any, Literal, TypedDict, cast
@@ -25,7 +24,7 @@ from langchain_core.language_models.chat_model_stream import AsyncChatModelStrea
 from langchain_protocol import Event, SubscribeParams

 from langgraph_sdk._async.http import HttpClient
-from langgraph_sdk._shared.utilities import _quote_path_param
+from langgraph_sdk._shared.utilities import _quote_path_param, _reconnect_delay
 from langgraph_sdk.schema import LangSmithTracing, QueryParamTypes
 from langgraph_sdk.stream.controller import _SeenEventIds
 from langgraph_sdk.stream.decoders import (
@@ -1691,11 +1690,13 @@ class AsyncThreadStream:

    async def _reconnect_sleep(self, attempt: int) -> None:
        """Sleep with exponential backoff and jitter for reconnect attempt `attempt`."""
-        base = self._shared_reconnect_backoff_base
-        cap = self._shared_reconnect_backoff_cap
-        delay = min(cap, base * (2**attempt))
-        jitter = random.uniform(0, delay * 0.25)
-        await asyncio.sleep(delay + jitter)
+        await asyncio.sleep(
+            _reconnect_delay(
+                attempt,
+                base=self._shared_reconnect_backoff_base,
+                cap=self._shared_reconnect_backoff_cap,
+            )
+        )

    async def _reconnect_shared_stream(self) -> bool:
        """Attempt to reopen the shared stream after a post-ready transport drop.
@@ -4,6 +4,7 @@ from __future__ import annotations

 import functools
 import os
+import random
 import re
 from collections.abc import Mapping
 from datetime import tzinfo
@@ -194,6 +195,15 @@ def _validate_reconnect_location(base_url: httpx.URL, location: str) -> str:
    return location


+def _reconnect_delay(attempt: int, *, base: float, cap: float) -> float:
+    """Seconds to wait before reconnect attempt `attempt` (0-based).
+
+    Exponential backoff capped at `cap`, plus up to 25% jitter.
+    """
+    delay = min(cap, base * (2**attempt))
+    return delay + random.uniform(0, delay * 0.25)
+
+
 def _provided_vals(d: Mapping[str, Any]) -> dict[str, Any]:
    return {k: v for k, v in d.items() if v is not None}

@@ -4,6 +4,7 @@ from __future__ import annotations

 import logging
 import sys
+import time
 import warnings
 from collections.abc import Callable, Iterator, Mapping
 from typing import Any, cast
@@ -13,9 +14,14 @@ import orjson

 from langgraph_sdk._shared.utilities import (
    _orjson_default,
+    _reconnect_delay,
    _validate_reconnect_location,
 )
-from langgraph_sdk.errors import _raise_for_status_typed
+from langgraph_sdk.errors import (
+    InternalServerError,
+    RateLimitError,
+    _raise_for_status_typed,
+)
 from langgraph_sdk.schema import QueryParamTypes, StreamPart
 from langgraph_sdk.sse import SSEDecoder, iter_lines_raw

@@ -146,7 +152,8 @@ class SyncHttpClient:
        on_response: Callable[[httpx.Response], None] | None = None,
        reconnect_limit: int = 5,
    ) -> Any:
-        """Send a request that automatically reconnects to Location header."""
+        """Send a request that reconnects to its Location header if the response is
+        cut off, with backoff between attempts."""
        request_headers, content = _encode_json(json)
        if headers:
            request_headers.update(headers)
@@ -176,13 +183,27 @@ class SyncHttpClient:
                    stacklevel=2,
                )
                r.close()
-                return self.request_reconnect(
-                    loc,
-                    "GET",
-                    headers=request_headers,
-                    # don't pass on_response so it's only called once
-                    reconnect_limit=reconnect_limit - 1,
-                )
+                for attempt in range(reconnect_limit):
+                    time.sleep(_reconnect_delay(attempt, base=1.0, cap=5.0))
+                    # A restart or a deploy looks like a refused connection, a
+                    # dropped body, a 429 or a 5xx; other errors won't go away.
+                    try:
+                        return self.request_reconnect(
+                            loc,
+                            "GET",
+                            headers=headers,
+                            # don't pass on_response so it's only called once
+                            reconnect_limit=0,
+                        )
+                    except httpx.HTTPStatusError as err:
+                        status = err.response.status_code
+                        if attempt == reconnect_limit - 1 or (
+                            status != 429 and status < 500
+                        ):
+                            raise
+                    except httpx.TransportError:
+                        if attempt == reconnect_limit - 1:
+                            raise

    def stream(
        self,
@@ -220,73 +241,105 @@ class SyncHttpClient:
                request_headers if reconnect_path is None else reconnect_headers
            )
            if last_event_id is not None:
+                current_headers = {
+                    key: value
+                    for key, value in current_headers.items()
+                    if key.lower() != "last-event-id"
+                }
+            if last_event_id:
                current_headers["Last-Event-ID"] = last_event_id
+            if reconnect_path is not None and not any(
+                key.lower() == "last-event-id" and value
+                for key, value in current_headers.items()
+            ):
+                warnings.warn(
+                    "Reconnecting without a Last-Event-ID, so events sent while "
+                    "disconnected are not replayed. Only runs created with "
+                    "stream_resumable=True send the event IDs a reconnect resumes "
+                    "from.",
+                    stacklevel=2,
+                )

            current_method = method if reconnect_path is None else "GET"
            current_content = content if reconnect_path is None else None
            current_params = params if reconnect_path is None else None

-            retry = False
-            with self.client.stream(
-                current_method,
-                reconnect_path or path,
-                headers=current_headers,
-                content=current_content,
-                params=current_params,
-            ) as res:
-                if reconnect_path is None and on_response:
-                    on_response(res)
-                # check status
-                _raise_for_status_typed(res)
-                # check content type
-                content_type = res.headers.get("content-type", "").partition(";")[0]
-                if "text/event-stream" not in content_type:
-                    raise httpx.TransportError(
-                        "Expected response header Content-Type to contain 'text/event-stream', "
-                        f"got {content_type!r}"
-                    )
+            reconnect_cause: httpx.HTTPError | None = None
+            try:
+                with self.client.stream(
+                    current_method,
+                    reconnect_path or path,
+                    headers=current_headers,
+                    content=current_content,
+                    params=current_params,
+                ) as res:
+                    if reconnect_path is None and on_response:
+                        on_response(res)
+                    # check status
+                    _raise_for_status_typed(res)
+                    # check content type
+                    content_type = res.headers.get("content-type", "").partition(";")[0]
+                    if "text/event-stream" not in content_type:
+                        raise httpx.TransportError(
+                            "Expected response header Content-Type to contain 'text/event-stream', "
+                            f"got {content_type!r}"
+                        )

-                reconnect_location = res.headers.get("location")
-                if reconnect_location:
-                    _validate_reconnect_location(
-                        self.client.base_url, reconnect_location
-                    )
-                    reconnect_path = reconnect_location
+                    reconnect_location = res.headers.get("location")
+                    if reconnect_location:
+                        _validate_reconnect_location(
+                            self.client.base_url, reconnect_location
+                        )
+                        reconnect_path = reconnect_location

-                decoder = SSEDecoder()
-                try:
-                    for line in iter_lines_raw(res):
-                        sse = decoder.decode(cast(bytes, line).rstrip(b"\n"))
-                        if sse is not None:
+                    decoder = SSEDecoder(last_event_id=last_event_id)
+                    try:
+                        for line in iter_lines_raw(res):
+                            sse = decoder.decode(cast(bytes, line).rstrip(b"\n"))
+                            if sse is not None:
+                                if decoder.last_event_id is not None:
+                                    last_event_id = decoder.last_event_id
+                                if sse.event or decoder.has_data:
+                                    # Progress since the last drop earns a fresh
+                                    # set of reconnect attempts.
+                                    reconnect_attempts = 0
+                                    yield sse
+                    except httpx.HTTPError as err:
+                        # httpx.TransportError inherits from HTTPError, so transient
+                        # disconnects during streaming land here.
+                        if reconnect_path is None:
+                            raise
+                        reconnect_cause = err
+                    else:
+                        if sse := decoder.decode(b""):
                            if decoder.last_event_id is not None:
                                last_event_id = decoder.last_event_id
-                            if sse.event or sse.data is not None:
+                            if sse.event or decoder.has_data:
+                                # See async stream implementation for rationale on
+                                # skipping empty flush events.
                                yield sse
-                except httpx.HTTPError:
-                    # httpx.TransportError inherits from HTTPError, so transient
-                    # disconnects during streaming land here.
-                    if reconnect_path is None:
-                        raise
-                    retry = True
-                else:
-                    if sse := decoder.decode(b""):
-                        if decoder.last_event_id is not None:
-                            last_event_id = decoder.last_event_id
-                        if sse.event or sse.data is not None:
-                            # See async stream implementation for rationale on
-                            # skipping empty flush events.
-                            yield sse
-            if retry:
+            except (httpx.TransportError, RateLimitError, InternalServerError) as err:
+                # A reconnect that can't open (connection refused, 429, 5xx or a
+                # non-SSE body) usually means the server is restarting or
+                # mid-deploy, so retry it like a drop. Before the server sends a
+                # Location there is nothing to reconnect to, so raise.
+                if reconnect_path is None:
+                    raise
+                reconnect_cause = err
+            if reconnect_cause is not None:
                reconnect_attempts += 1
                if reconnect_attempts > max_reconnect_attempts:
                    raise httpx.TransportError(
                        "Exceeded maximum SSE reconnection attempts"
-                    )
+                    ) from reconnect_cause
+                time.sleep(_reconnect_delay(reconnect_attempts - 1, base=1.0, cap=5.0))
                continue
            break


-def _encode_json(json: Any) -> tuple[dict[str, str], bytes]:
+def _encode_json(json: Any) -> tuple[dict[str, str], bytes | None]:
+    if json is None:
+        return {}, None
    body = orjson.dumps(
        json,
        _orjson_default,
@@ -80,6 +80,7 @@ class SyncRunsClient:
        command: Command | None = None,
        stream_mode: StreamMode | Sequence[StreamMode] = "values",
        stream_subgraphs: bool = False,
+        stream_resumable: bool = False,
        metadata: Mapping[str, Any] | None = None,
        config: Config | None = None,
        context: Context | None = None,
@@ -98,6 +99,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v1"] = "v1",
    ) -> Iterator[StreamPart]: ...

@@ -111,6 +113,7 @@ class SyncRunsClient:
        command: Command | None = None,
        stream_mode: StreamMode | Sequence[StreamMode] = "values",
        stream_subgraphs: bool = False,
+        stream_resumable: bool = False,
        metadata: Mapping[str, Any] | None = None,
        config: Config | None = None,
        context: Context | None = None,
@@ -129,6 +132,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v2"],
    ) -> Iterator[StreamPartV2]: ...

@@ -159,6 +163,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v1"] = "v1",
    ) -> Iterator[StreamPart]: ...

@@ -189,6 +194,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
        version: Literal["v2"],
    ) -> Iterator[StreamPartV2]: ...

@@ -380,6 +386,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> Run: ...

    @overload
@@ -409,6 +416,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> Run: ...

    def create(
@@ -642,6 +650,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> builtins.list[dict] | dict[str, Any]: ...

    @overload
@@ -668,6 +677,7 @@ class SyncRunsClient:
        headers: Mapping[str, str] | None = None,
        params: QueryParamTypes | None = None,
        on_run_created: Callable[[RunCreateMetadata], None] | None = None,
+        durability: Durability | None = None,
    ) -> builtins.list[dict] | dict[str, Any]: ...

    def wait(
@@ -815,7 +825,6 @@ class SyncRunsClient:
            "checkpoint_during": checkpoint_during,
            "on_completion": on_completion,
            "after_seconds": after_seconds,
-            "raise_error": raise_error,
            "durability": durability,
            "langsmith_tracer": langsmith_tracing,
        }
@@ -830,7 +839,7 @@ class SyncRunsClient:
            if thread_id is not None
            else "/runs/wait"
        )
-        return self.http.request_reconnect(
+        response = self.http.request_reconnect(
            endpoint,
            "POST",
            json={k: v for k, v in payload.items() if v is not None},
@@ -838,6 +847,16 @@ class SyncRunsClient:
            headers=headers,
            on_response=on_response if on_run_created else None,
        )
+        if (
+            raise_error
+            and isinstance(response, dict)
+            and "__error__" in response
+            and isinstance(response["__error__"], dict)
+        ):
+            raise Exception(
+                f"{response['__error__'].get('error')}: {response['__error__'].get('message')}"
+            )
+        return response

    def list(
        self,
@@ -811,7 +811,7 @@ class _On:
    def __call__(
        self,
        *,
-        resources: str | Sequence[str],
+        resources: str | Sequence[str] | None = None,
        actions: str | Sequence[str] | None = None,
    ) -> Callable[[AHO], AHO]: ...

@@ -0,0 +1,155 @@
+"""Custom run priority for LangGraph.
+
+.. warning::
+    This API is in beta and may change in future versions.
+
+This module lets developers decide the priority settings of each thread with
+an async function that the server calls at the first run on a thread with no
+settings. The function is registered on an object that `langgraph.json` names,
+and runs server-side.
+"""
+
+from __future__ import annotations
+
+import functools
+import inspect
+import typing
+import warnings
+from collections.abc import Awaitable, Callable
+
+from langgraph_sdk.encryption import DuplicateHandlerError, LangGraphBetaWarning
+from langgraph_sdk.priority import types
+
+_AssignHandler = Callable[
+    [types.PriorityContext], Awaitable[types.PrioritySettings | None]
+]
+_AssignHandlerT = typing.TypeVar("_AssignHandlerT", bound=_AssignHandler)
+
+
+@functools.lru_cache(maxsize=1)
+def _warn_priority_beta() -> None:
+    warnings.warn(
+        "The Priority API is in beta and may change in future versions.",
+        LangGraphBetaWarning,
+        stacklevel=4,
+    )
+
+
+def _validate_handler(fn: typing.Callable) -> None:
+    """Validate that the assign handler is async and takes one parameter.
+
+    Raises:
+        TypeError: If the handler is not an async function or has wrong parameter count
+    """
+    if not inspect.iscoroutinefunction(fn):
+        raise TypeError(
+            f"Priority assign handler must be an async function, got {type(fn)}"
+        )
+
+    sig = inspect.signature(fn)
+    params = [
+        p
+        for p in sig.parameters.values()
+        if p.kind in (p.POSITIONAL_ONLY, p.POSITIONAL_OR_KEYWORD)
+    ]
+    if len(params) != 1:
+        raise TypeError(
+            "Priority assign handler must accept exactly 1 parameter (ctx), "
+            f"got {len(params)}"
+        )
+
+
+class Priority:
+    """Decide the priority settings of your LangGraph application's threads.
+
+    .. warning::
+        This API is in beta and may change in future versions.
+
+    To use, create a separate Python file and add the path to the file to your
+    LangGraph API configuration file (`langgraph.json`). Within that file, create
+    an instance of the Priority class and register an assign function with
+    `@priority.assign`.
+
+    Example `langgraph.json` file:
+
+    ```json
+    {
+      "dependencies": ["."],
+      "graphs": {
+        "agent": "./my_agent/agent.py:graph"
+      },
+      "env": ".env",
+      "priority": {
+        "path": "./priority.py:priority"
+      }
+    }
+    ```
+
+    The server calls the assign function at the first run on a thread with no
+    settings, which can come long after the thread was created, and stores the
+    settings it returns on the thread. Later runs on the thread use the stored
+    settings. The function can be called more than once for one thread, for
+    example after it raises, and only one result is stored, so keep it free of
+    side effects.
+
+    ???+ example "Basic Usage"
+
+        ```python
+        from langgraph_sdk import Priority, PriorityContext, PrioritySettings
+
+        priority = Priority()
+
+        @priority.assign
+        async def assign(ctx: PriorityContext) -> PrioritySettings | None:
+            return {
+                "priority": 3 if "premium" in ctx.permissions else 0,
+                # Stored unencrypted, so an opaque ID rather than personal data.
+                "fairness_group": ctx.user["org_id"],
+            }
+        ```
+    """
+
+    __slots__ = ("_assign_handler",)
+
+    types = types
+    """Reference to priority type definitions.
+
+    Provides access to the type definitions used by the priority system:
+    PriorityContext and PrioritySettings.
+    """
+
+    def __init__(self) -> None:
+        """Initialize the Priority instance."""
+        _warn_priority_beta()
+        self._assign_handler: _AssignHandler | None = None
+
+    def assign(self, fn: _AssignHandlerT) -> _AssignHandlerT:
+        """Register the function that assigns a thread's priority settings.
+
+        The function receives a `PriorityContext` and returns `PrioritySettings`,
+        or `None` for the defaults.
+
+        Example:
+            ```python
+            @priority.assign
+            async def assign(ctx: PriorityContext) -> PrioritySettings | None:
+                if ctx.cron_id is not None:
+                    return {"priority": -2}
+                return None
+            ```
+
+        Args:
+            fn: The assign handler function
+
+        Returns:
+            The registered handler function
+
+        Raises:
+            DuplicateHandlerError: If an assign handler is already registered
+            TypeError: If the handler has an invalid signature
+        """
+        if self._assign_handler is not None:
+            raise DuplicateHandlerError("Priority assign handler already registered")
+        _validate_handler(fn)
+        self._assign_handler = fn
+        return fn
@@ -0,0 +1,67 @@
+"""Priority types for LangGraph.
+
+This module defines the context the server passes to the function registered
+with `@priority.assign`, and the settings that function returns.
+"""
+
+from __future__ import annotations
+
+import typing
+from collections.abc import Sequence
+from dataclasses import dataclass
+
+from langgraph_sdk.auth.types import BaseUser
+
+
+@dataclass(frozen=True, slots=True)
+class PriorityContext:
+    """Context passed to the `@priority.assign` function at the first run on a thread with no settings."""
+
+    user: BaseUser
+    """The user from the deployment's auth, custom auth or LangSmith auth. Without auth, `user.identity` is empty."""
+
+    permissions: Sequence[str]
+    """The user's permissions from the same auth. Without auth, it's empty."""
+
+    thread_id: str | None
+    """The thread's ID, or `None` when the run creates its thread without being given an ID."""
+
+    thread_metadata: dict[str, typing.Any]
+    """The thread's metadata."""
+
+    assistant_id: str
+    """The ID of the assistant the run uses."""
+
+    graph_id: str
+    """The ID of the graph the assistant runs."""
+
+    assistant_metadata: dict[str, typing.Any]
+    """The assistant's metadata."""
+
+    run_metadata: dict[str, typing.Any]
+    """The run's metadata."""
+
+    cron_id: str | None
+    """The ID of the cron that started the run, or `None` when no cron started it."""
+
+
+class PrioritySettings(typing.TypedDict, total=False):
+    """Priority settings returned by the `@priority.assign` function.
+
+    Every key is optional: a missing key or a key set to `None` takes its
+    default, and returning `None` instead of a dict gives every key its default.
+    A value outside its range, or an empty `fairness_group`, is an error.
+    """
+
+    priority: int | None
+    """An integer from -5 to 5, where 5 runs first and -5 runs last. Defaults to 0."""
+
+    fairness_group: str | None
+    """The thread's fairness group, a non-empty string. Defaults to `"default"`.
+
+    The server stores it unencrypted, even when the deployment uses custom
+    encryption, so use an opaque ID rather than personal data.
+    """
+
+    fairness_weight: int | None
+    """The weight of the thread's fairness group, an integer from 1 to 100. Defaults to 1."""
@@ -30,18 +30,14 @@ class BytesLineDecoder:
        # See https://docs.python.org/3/glossary.html#term-universal-newlines
        NEWLINE_CHARS = b"\n\r"

-        # We always push a trailing `\r` into the next decode iteration.
-        if self.trailing_cr:
-            text = b"\r" + text
-            self.trailing_cr = False
-        if text.endswith(b"\r"):
-            self.trailing_cr = True
-            text = text[:-1]
+        if not text:
+            return []
+        if self.trailing_cr and text.startswith(b"\n"):
+            text = text[1:]
+        self.trailing_cr = text.endswith(b"\r")

        if not text:
-            # NOTE: the edge case input of empty text doesn't occur in practice,
-            # because other httpx internals filter out this value
-            return []  # pragma: no cover
+            return []

        trailing_newline = text[-1] in NEWLINE_CHARS
        lines = cast(list[BytesLike], text.splitlines())
@@ -66,7 +62,7 @@ class BytesLineDecoder:
        return lines

    def flush(self) -> list[BytesLike]:
-        if not self.buffer and not self.trailing_cr:
+        if not self.buffer:
            return []

        lines: list[BytesLike] = [self.buffer]
@@ -76,40 +72,66 @@ class BytesLineDecoder:


 class SSEDecoder:
-    def __init__(self) -> None:
+    def __init__(self, *, last_event_id: str | None = None) -> None:
        self._event = ""
        self._data = bytearray()
-        self._last_event_id = ""
+        self._last_event_id = last_event_id
        self._retry: int | None = None
+        self._at_stream_start = True
+        self._has_data = False
+        self._id_changed = False
+
+    @property
+    def has_data(self) -> bool:
+        """Whether the last decoded block contained a data field."""
+        return self._has_data

    @property
    def last_event_id(self) -> str | None:
-        """Return the last event identifier that was seen."""
+        """Return the last event identifier that was seen.

-        return self._last_event_id or None
+        `None` until an ID is supplied or received, and `""` after an empty
+        one, which resets the reconnect cursor.
+        """
+
+        return self._last_event_id

    def decode(self, line: bytes) -> StreamPart | None:
        # See: https://html.spec.whatwg.org/multipage/server-sent-events.html#event-stream-interpretation

+        if self._at_stream_start:
+            self._at_stream_start = False
+            line = line.removeprefix(b"\xef\xbb\xbf")
+
        if not line:
+            self._has_data = bool(self._data)
            if (
                not self._event
                and not self._data
-                and not self._last_event_id
+                and not self._id_changed
                and self._retry is None
            ):
                return None

+            body = self._data[:-1]
+            data = None
+            if body and not body.isspace():
+                try:
+                    data = orjson.loads(body)
+                except orjson.JSONDecodeError:
+                    data = orjson.loads(body.decode(errors="replace"))
+
            sse = StreamPart(
                event=self._event,
-                data=orjson.loads(self._data) if self._data else None,  # ty: ignore[invalid-argument-type]
-                id=self.last_event_id,
+                data=data,  # ty: ignore[invalid-argument-type]
+                id=self._last_event_id or None,
            )

            # NOTE: as per the SSE spec, do not reset last_event_id.
            self._event = ""
            self._data = bytearray()
            self._retry = None
+            self._id_changed = False

            return sse

@@ -122,14 +144,16 @@ class SSEDecoder:
            value = value[1:]

        if fieldname == b"event":
-            self._event = value.decode()
+            self._event = value.decode(errors="replace")
        elif fieldname == b"data":
            self._data.extend(value)
+            self._data.extend(b"\n")
        elif fieldname == b"id":
            if b"\0" in value:
                pass
            else:
-                self._last_event_id = value.decode()
+                self._last_event_id = value.decode(errors="replace")
+                self._id_changed = True
        elif fieldname == b"retry":
            with contextlib.suppress(TypeError, ValueError):
                self._retry = int(value)
@@ -15,7 +15,6 @@ from __future__ import annotations
 import asyncio
 import contextlib
 import logging
-import random
 from collections import OrderedDict
 from collections.abc import AsyncGenerator, AsyncIterator, Awaitable, Callable
 from dataclasses import dataclass, field
@@ -23,6 +22,7 @@ from typing import Any

 from langchain_protocol import Event, SubscribeParams

+from langgraph_sdk._shared.utilities import _reconnect_delay
 from langgraph_sdk.stream.transport import AsyncProtocolTransport, EventStreamHandle

 # ---------------------------------------------------------------------------
@@ -275,11 +275,13 @@ class StreamController:

    async def _reconnect_sleep(self, attempt: int) -> None:
        """Sleep with exponential backoff and jitter for reconnect attempt *attempt*."""
-        base = self._reconnect_backoff_base
-        cap = self._reconnect_backoff_cap
-        delay = min(cap, base * (2**attempt))
-        jitter = random.uniform(0, delay * 0.25)
-        await asyncio.sleep(delay + jitter)
+        await asyncio.sleep(
+            _reconnect_delay(
+                attempt,
+                base=self._reconnect_backoff_base,
+                cap=self._reconnect_backoff_cap,
+            )
+        )

    async def _reconnect_shared_stream(self) -> bool:
        """Attempt to reopen the shared stream after a transport drop.
@@ -4,7 +4,6 @@ from __future__ import annotations

 import contextlib
 import logging
-import random
 import threading
 import time
 from dataclasses import dataclass, field
@@ -13,6 +12,7 @@ from typing import Any

 from langchain_protocol import Event, SubscribeParams

+from langgraph_sdk._shared.utilities import _reconnect_delay
 from langgraph_sdk.stream.errors import (
    is_non_retryable_stream_error,
    stream_error_frame,
@@ -282,11 +282,13 @@ class SyncStreamController:

    def _reconnect_sleep(self, attempt: int) -> None:
        """Sleep with exponential backoff + jitter before a reconnect attempt."""
-        base = self._reconnect_backoff_base
-        cap = self._reconnect_backoff_cap
-        delay = min(cap, base * (2**attempt))
-        jitter = random.uniform(0, delay * 0.25)
-        time.sleep(delay + jitter)
+        time.sleep(
+            _reconnect_delay(
+                attempt,
+                base=self._reconnect_backoff_base,
+                cap=self._reconnect_backoff_cap,
+            )
+        )

    def _reconnect_shared_stream(self) -> bool:
        """Attempt to reopen the shared stream after a transport drop.
@@ -16,7 +16,7 @@ dependencies = [
    "orjson>=3.11.5",
    "langchain-protocol>=0.0.15",
    "langchain-core>=1.4.0,<2",
-    "websockets>=14,<17",
+    "websockets>=14",
 ]

 [tool.hatch.version]
@@ -0,0 +1,17 @@
+import asyncio
+import time
+
+import pytest
+
+
+@pytest.fixture
+def no_sleep(monkeypatch):
+    """Record backoff sleeps without actually waiting."""
+    sleeps: list[float] = []
+
+    async def fake_async_sleep(d):
+        sleeps.append(d)
+
+    monkeypatch.setattr(asyncio, "sleep", fake_async_sleep)
+    monkeypatch.setattr(time, "sleep", lambda d: sleeps.append(d))
+    return sleeps
@@ -1,7 +1,9 @@
 from __future__ import annotations

+import ast
 import inspect
 import re
+from pathlib import Path

 import pytest

@@ -128,3 +130,45 @@ def test_sync_api_matches_async(async_cls, sync_cls):
            f"Return annotation mismatch for {async_cls.__name__}.{name}: "
            f"async={a_ret}, sync={s_ret}"
        )
+
+
+# Stateless overloads (`thread_id: None`) leave out the thread-only parameters,
+# and stateful ones (`thread_id: str`) leave out `on_completion`, which only
+# applies to stateless runs.
+_RUN_OVERLOAD_OMISSIONS = {
+    "None": {"checkpoint", "checkpoint_id", "multitask_strategy"},
+    "str": {"on_completion"},
+}
+
+
+@pytest.mark.parametrize("client_cls", [RunsClient, SyncRunsClient])
+def test_run_overloads_accept_implementation_params(client_cls) -> None:
+    tree = ast.parse(Path(inspect.getfile(client_cls)).read_text())
+    cls = next(
+        node
+        for node in tree.body
+        if isinstance(node, ast.ClassDef) and node.name == client_cls.__name__
+    )
+    defs = [
+        node
+        for node in cls.body
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))
+        and node.name in {"stream", "create", "wait"}
+    ]
+    impl_params = {
+        node.name: {arg.arg for arg in node.args.kwonlyargs}
+        for node in defs
+        if not node.decorator_list
+    }
+    gaps = []
+    for stub in defs:
+        if not stub.decorator_list:
+            continue
+        thread_id = next(arg for arg in stub.args.args if arg.arg == "thread_id")
+        assert thread_id.annotation is not None
+        kind = ast.unparse(thread_id.annotation)
+        expected = impl_params[stub.name] - _RUN_OVERLOAD_OMISSIONS[kind]
+        if missing := expected - {arg.arg for arg in stub.args.kwonlyargs}:
+            gaps.append(f"{stub.name}(thread_id: {kind}) line {stub.lineno}: {missing}")
+
+    assert not gaps, gaps
@@ -1,8 +1,49 @@
+import ast
+import inspect
+
 import pytest

 from langgraph_sdk import Auth


+@pytest.mark.parametrize(
+    "kwargs",
+    [{}, {"actions": "read"}, {"resources": None}],
+)
+def test_global_handler_factory_accepts_optional_resources(kwargs) -> None:
+    auth = Auth()
+    tree = ast.parse(inspect.getsource(type(auth.on)))
+    cls = tree.body[0]
+    assert isinstance(cls, ast.ClassDef)
+    factory_overload = next(
+        node
+        for node in cls.body
+        if isinstance(node, ast.FunctionDef)
+        and node.name == "__call__"
+        and node.decorator_list
+        and any(arg.arg == "resources" for arg in node.args.kwonlyargs)
+    )
+    resources = next(
+        arg for arg in factory_overload.args.kwonlyargs if arg.arg == "resources"
+    )
+    assert resources.annotation is not None
+    assert ast.unparse(resources.annotation) == "str | Sequence[str] | None"
+    default = factory_overload.args.kw_defaults[0]
+    assert isinstance(default, ast.Constant) and default.value is None
+
+    @auth.on(**kwargs)
+    async def handler(ctx, value):
+        del ctx, value
+        return None
+
+    if kwargs.get("actions"):
+        assert auth._handlers == {("*", "read"): [handler]}
+        assert auth._global_handlers == []
+    else:
+        assert auth._global_handlers == [handler]
+        assert auth._handlers == {}
+
+
 def test_handler_multiple_resources_and_actions() -> None:
    auth = Auth()

@@ -1,10 +1,11 @@
 from __future__ import annotations

-from collections.abc import Iterator, Sequence
+from collections.abc import Callable, Iterator, Sequence
 from pathlib import Path
 from typing import Any

 import httpx
+import orjson
 import pytest
 from typing_extensions import assert_type

@@ -12,6 +13,7 @@ from langgraph_sdk._async.runs import _wrap_stream_v2
 from langgraph_sdk._shared.utilities import _sse_to_v2_dict
 from langgraph_sdk._sync.runs import _wrap_stream_v2_sync
 from langgraph_sdk.client import HttpClient, SyncHttpClient
+from langgraph_sdk.errors import InternalServerError, NotFoundError
 from langgraph_sdk.schema import (
    CheckpointPayload,
    CheckpointsStreamPart,
@@ -110,6 +112,129 @@ def test_stream_sse():
        assert len(parts) == 79


+def test_sse_decoder_joins_data_lines_with_a_newline():
+    decoder = SSEDecoder()
+    for line in (b"data: [1,", b"data: 2]"):
+        decoder.decode(line)
+    assert decoder.decode(b"") == StreamPart(event="", data=[1, 2], id=None)  # ty: ignore[invalid-argument-type]
+
+    for line in (b"data: [1", b"data: 0]"):
+        decoder.decode(line)
+    with pytest.raises(orjson.JSONDecodeError):
+        decoder.decode(b"")
+
+
+def test_sse_decoder_strips_a_leading_bom():
+    decoder = SSEDecoder()
+    for line in (b"\xef\xbb\xbfevent: values", b"data: 1"):
+        decoder.decode(line)
+    assert decoder.decode(b"") == StreamPart(event="values", data=1, id=None)  # ty: ignore[invalid-argument-type]
+
+
+def test_control_blocks_do_not_repeat_after_a_cursor_update():
+    decoder = SSEDecoder()
+    decoder.decode(b"id: cursor")
+    assert decoder.decode(b"") == StreamPart(event="", data=None, id="cursor")  # ty: ignore[invalid-argument-type]
+    decoder.decode(b": heartbeat")
+    assert decoder.decode(b"") is None
+    assert decoder.decode(b"") is None
+    decoder.decode(b"id:")
+    assert decoder.decode(b"") == StreamPart(event="", data=None, id=None)  # ty: ignore[invalid-argument-type]
+    assert decoder.last_event_id == ""
+    assert decoder.decode(b"") is None
+
+
+def test_cr_lines_are_emitted_without_waiting_for_another_chunk():
+    decoder = BytesLineDecoder()
+    assert decoder.decode(b"data: 1\r\r") == [b"data: 1", b""]
+    assert decoder.decode(b"\n") == []
+    assert decoder.decode(b"data: 2\r") == [b"data: 2"]
+    assert decoder.decode(b"") == []
+    assert decoder.decode(b"\n\r") == [b""]
+    assert decoder.flush() == []
+
+
+def test_empty_data_lines_preserve_their_newlines():
+    decoder = SSEDecoder()
+    for line in (b"data", b"data: 1", b"data:"):
+        decoder.decode(line)
+    assert decoder._data == b"\n1\n\n"
+    assert decoder.decode(b"") == StreamPart(event="", data=1, id=None)  # ty: ignore[invalid-argument-type]
+    decoder.decode(b"data:")
+    decoder.decode(b"data:")
+    assert decoder.decode(b"") == StreamPart(event="", data=None, id=None)  # ty: ignore[invalid-argument-type]
+    assert decoder.has_data
+    assert decoder.decode(b"") is None
+    assert not decoder.has_data
+
+
+def test_malformed_utf8_is_replaced_in_fields_and_json_strings():
+    decoder = SSEDecoder()
+    for line in (b"event: \xff", b"id: \xff", b'data: "\xff"'):
+        decoder.decode(line)
+    assert decoder.decode(b"") == StreamPart(event="\ufffd", data="\ufffd", id="\ufffd")  # ty: ignore[invalid-argument-type]
+
+
+@pytest.mark.parametrize("asynchronous", [False, True])
+async def test_null_and_empty_data_are_not_dropped(asynchronous: bool):
+    def handler(_request: httpx.Request) -> httpx.Response:
+        return httpx.Response(
+            200,
+            headers={"Content-Type": "text/event-stream"},
+            content=b"data: null\n\ndata:\n\n",
+        )
+
+    transport = httpx.MockTransport(handler)
+    if asynchronous:
+        async with httpx.AsyncClient(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            parts = [part async for part in HttpClient(client).stream("/stream", "GET")]
+    else:
+        with httpx.Client(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            parts = list(SyncHttpClient(client).stream("/stream", "GET"))
+    assert parts == [StreamPart(event="", data=None), StreamPart(event="", data=None)]  # ty: ignore[invalid-argument-type]
+
+
+@pytest.mark.parametrize("asynchronous", [False, True])
+async def test_cursor_is_inherited_after_reconnect(asynchronous: bool):
+    requests: list[httpx.Request] = []
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        requests.append(request)
+        if len(requests) == 1:
+            stream_cls = AsyncListByteStream if asynchronous else ListByteStream
+            return httpx.Response(
+                200,
+                headers={"Content-Type": "text/event-stream", "Location": "/reconnect"},
+                stream=stream_cls(
+                    [b"id: cursor\nevent: values\ndata: 1\n\n"],
+                    httpx.RemoteProtocolError("incomplete chunked read"),
+                ),
+            )
+        assert (b"Last-Event-ID", b"cursor") in request.headers.raw
+        return httpx.Response(
+            200,
+            headers={"Content-Type": "text/event-stream"},
+            content=b"event: values\ndata: 2\n\n",
+        )
+
+    transport = httpx.MockTransport(handler)
+    if asynchronous:
+        async with httpx.AsyncClient(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            parts = [part async for part in HttpClient(client).stream("/stream", "GET")]
+    else:
+        with httpx.Client(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            parts = list(SyncHttpClient(client).stream("/stream", "GET"))
+    assert [part.id for part in parts] == ["cursor", "cursor"]
+
+
 # --- HTTP client streaming ---


@@ -156,6 +281,7 @@ def test_sync_http_client_stream_flushes_trailing_event():
    assert parts == [StreamPart(event="foo", data={"bar": 1})]


+@pytest.mark.usefixtures("no_sleep")
 def test_sync_http_client_stream_recovers_after_disconnect():
    reconnect_path = "/reconnect"
    first_chunks = [
@@ -221,6 +347,7 @@ def test_sync_http_client_stream_recovers_after_disconnect():


@pytest.mark.asyncio
+@pytest.mark.usefixtures("no_sleep")
 async def test_http_client_stream_recovers_after_disconnect():
    reconnect_path = "/reconnect"
    first_chunks = [
@@ -292,6 +419,431 @@ async def test_http_client_stream_recovers_after_disconnect():
    ]


+def _drop_after(chunks: list[bytes], stream_cls: type, reconnects: list[httpx.Request]):
+    def handler(request: httpx.Request) -> httpx.Response:
+        if request.method == "POST":
+            return httpx.Response(
+                200,
+                headers={"Content-Type": "text/event-stream", "Location": "/reconnect"},
+                stream=stream_cls(
+                    chunks, httpx.RemoteProtocolError("incomplete chunked read")
+                ),
+            )
+        reconnects.append(request)
+        return httpx.Response(
+            200,
+            headers={"Content-Type": "text/event-stream"},
+            stream=stream_cls([b"event: end\ndata: null\n\n"]),
+        )
+
+    return handler
+
+
+def _sync_reconnect(
+    chunks: list[bytes], headers: dict[str, str] | None
+) -> httpx.Request:
+    reconnects: list[httpx.Request] = []
+    transport = httpx.MockTransport(_drop_after(chunks, ListByteStream, reconnects))
+    with httpx.Client(transport=transport, base_url="https://example.com") as client:
+        list(SyncHttpClient(client).stream("/stream", "POST", json={}, headers=headers))
+    return reconnects[0]
+
+
+async def _async_reconnect(
+    chunks: list[bytes], headers: dict[str, str] | None
+) -> httpx.Request:
+    reconnects: list[httpx.Request] = []
+    transport = httpx.MockTransport(
+        _drop_after(chunks, AsyncListByteStream, reconnects)
+    )
+    async with httpx.AsyncClient(
+        transport=transport, base_url="https://example.com"
+    ) as client:
+        [
+            part
+            async for part in HttpClient(client).stream(
+                "/stream", "POST", json={}, headers=headers
+            )
+        ]
+    return reconnects[0]
+
+
+_EMPTY_IDS = pytest.mark.parametrize(
+    "empty_id",
+    [b"id:\nevent: values\ndata: 2\n\n", b"id:\n\n"],
+    ids=["with-an-event", "alone"],
+)
+_CALLER_CURSORS = pytest.mark.parametrize(
+    "headers",
+    [None, {"Last-Event-ID": "5"}, {"last-event-id": "5"}],
+    ids=["no-caller-cursor", "caller-cursor", "lowercase-caller-cursor"],
+)
+
+
+@_EMPTY_IDS
+@_CALLER_CURSORS
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_stream_reconnects_without_a_cursor_after_an_empty_id(
+    empty_id: bytes, headers: dict[str, str] | None
+):
+    with pytest.warns(UserWarning, match="Last-Event-ID"):
+        reconnect = _sync_reconnect(
+            [b"id: 1\nevent: values\ndata: 1\n\n", empty_id], headers
+        )
+
+    assert "last-event-id" not in reconnect.headers
+
+
+@_EMPTY_IDS
+@_CALLER_CURSORS
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_stream_reconnects_without_a_cursor_after_an_empty_id(
+    empty_id: bytes, headers: dict[str, str] | None
+):
+    with pytest.warns(UserWarning, match="Last-Event-ID"):
+        reconnect = await _async_reconnect(
+            [b"id: 1\nevent: values\ndata: 1\n\n", empty_id], headers
+        )
+
+    assert "last-event-id" not in reconnect.headers
+
+
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_stream_id_replaces_a_lowercase_caller_cursor():
+    reconnect = _sync_reconnect(
+        [b"id: 1\nevent: values\ndata: 1\n\n"], {"last-event-id": "5"}
+    )
+
+    assert reconnect.headers.get_list("last-event-id") == ["1"]
+
+
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_stream_id_replaces_a_lowercase_caller_cursor():
+    reconnect = await _async_reconnect(
+        [b"id: 1\nevent: values\ndata: 1\n\n"], {"last-event-id": "5"}
+    )
+
+    assert reconnect.headers.get_list("last-event-id") == ["1"]
+
+
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_stream_keeps_the_caller_cursor_until_the_stream_sends_an_id():
+    reconnect = _sync_reconnect([b"event: values\ndata: 1\n\n"], {"Last-Event-ID": "5"})
+
+    assert reconnect.headers.get_list("last-event-id") == ["5"]
+
+
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_stream_keeps_the_caller_cursor_until_the_stream_sends_an_id():
+    reconnect = await _async_reconnect(
+        [b"event: values\ndata: 1\n\n"], {"Last-Event-ID": "5"}
+    )
+
+    assert reconnect.headers.get_list("last-event-id") == ["5"]
+
+
+# A reconnect that sends no Last-Event-ID can't replay what was sent while it was
+# disconnected, so it warns.
+_RECONNECT_CURSORS = pytest.mark.parametrize(
+    ("chunks", "headers", "warns"),
+    [
+        ([b"event: values\ndata: 1\n\n"], None, True),
+        ([b"id: 1\nevent: values\ndata: 1\n\n"], None, False),
+        ([b"event: values\ndata: 1\n\n"], {"Last-Event-ID": "5"}, False),
+    ],
+    ids=["no-cursor", "stream-cursor", "caller-cursor"],
+)
+
+
+@_RECONNECT_CURSORS
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_reconnect_warns_only_without_a_cursor(
+    chunks: list[bytes],
+    headers: dict[str, str] | None,
+    warns: bool,
+    recwarn: pytest.WarningsRecorder,
+):
+    _sync_reconnect(chunks, headers)
+
+    assert any("Last-Event-ID" in str(w.message) for w in recwarn) is warns
+
+
+@_RECONNECT_CURSORS
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_reconnect_warns_only_without_a_cursor(
+    chunks: list[bytes],
+    headers: dict[str, str] | None,
+    warns: bool,
+    recwarn: pytest.WarningsRecorder,
+):
+    await _async_reconnect(chunks, headers)
+
+    assert any("Last-Event-ID" in str(w.message) for w in recwarn) is warns
+
+
+# --- reconnects survive a server restart or a deploy ---
+
+SSE = {"Content-Type": "text/event-stream"}
+DROP = httpx.RemoteProtocolError("incomplete chunked read")
+
+
+def _event(i: int) -> bytes:
+    return f"id: {i}\nevent: values\ndata: {i}\n\n".encode()
+
+
+def _scripted(steps: list[Callable[[httpx.Request], httpx.Response]]):
+    """Handler that plays one step per request and records each request."""
+    seen: list[httpx.Request] = []
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        seen.append(request)
+        if len(seen) > len(steps):
+            raise AssertionError("unexpected request")
+        return steps[len(seen) - 1](request)
+
+    return handler, seen
+
+
+def _ok(stream_cls, chunks, *, drop=False, location=None):
+    headers = dict(SSE)
+    if location:
+        headers["Location"] = location
+
+    def step(_request):
+        return httpx.Response(
+            200, headers=headers, stream=stream_cls(chunks, DROP if drop else None)
+        )
+
+    return step
+
+
+def _status(code):
+    return lambda request: httpx.Response(
+        code, json={"detail": "upstream"}, request=request
+    )
+
+
+def _connect_error(request):
+    raise httpx.ConnectError("[Errno 61] Connection refused", request=request)
+
+
+def _not_sse(request):
+    return httpx.Response(
+        200, headers={"Content-Type": "text/html"}, text="<html>", request=request
+    )
+
+
+def _transient_steps(stream_cls, failure):
+    """POST drops after event 1; first reconnect fails; second succeeds."""
+    return [
+        _ok(stream_cls, [_event(1)], drop=True, location="/reconnect"),
+        failure,
+        _ok(stream_cls, [_event(2), b"event: end\ndata: null\n\n"]),
+    ]
+
+
+def _long_stream_steps(stream_cls, drops):
+    """POST + `drops` reconnects, each delivering one event then dropping."""
+    steps = [_ok(stream_cls, [_event(0)], drop=True, location="/reconnect")]
+    for i in range(1, drops):
+        steps.append(_ok(stream_cls, [_event(i)], drop=True))
+    steps.append(_ok(stream_cls, [_event(drops), b"event: end\ndata: null\n\n"]))
+    return steps
+
+
+def _run_sync(steps):
+    handler, seen = _scripted(steps)
+    with httpx.Client(
+        transport=httpx.MockTransport(handler), base_url="https://example.com"
+    ) as client:
+        parts = list(SyncHttpClient(client).stream("/stream", "POST", json={}))
+    return parts, seen
+
+
+async def _run_async(steps):
+    sync_handler, seen = _scripted(steps)
+
+    async def handler(request):
+        return sync_handler(request)
+
+    async with httpx.AsyncClient(
+        transport=httpx.MockTransport(handler), base_url="https://example.com"
+    ) as client:
+        parts = [p async for p in HttpClient(client).stream("/stream", "POST", json={})]
+    return parts, seen
+
+
+FAILURES = {
+    "429": _status(429),
+    "502": _status(502),
+    "503": _status(503),
+    "connect_error": _connect_error,
+    "not_sse": _not_sse,
+}
+
+
+# A reconnect that hits a 429 / 502 / 503 / connect error / non-SSE body, then
+# succeeds, should recover.
+
+
+@pytest.mark.usefixtures("no_sleep")
+@pytest.mark.parametrize("failure", FAILURES.values(), ids=FAILURES.keys())
+def test_sync_reconnect_recovers_from_transient_failure(failure):
+    parts, seen = _run_sync(_transient_steps(ListByteStream, failure))
+    assert [p.event for p in parts] == ["values", "values", "end"]
+    assert len(seen) == 3
+
+
+@pytest.mark.usefixtures("no_sleep")
+@pytest.mark.parametrize("failure", FAILURES.values(), ids=FAILURES.keys())
+async def test_async_reconnect_recovers_from_transient_failure(failure):
+    parts, seen = await _run_async(_transient_steps(AsyncListByteStream, failure))
+    assert [p.event for p in parts] == ["values", "values", "end"]
+    assert len(seen) == 3
+
+
+# Any other 4xx on a reconnect won't go away by retrying, so it raises.
+
+
+def _client_error_steps(stream_cls):
+    return [
+        _ok(stream_cls, [_event(1)], drop=True, location="/reconnect"),
+        _status(404),
+    ]
+
+
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_reconnect_does_not_retry_a_client_error():
+    with pytest.raises(NotFoundError):
+        _run_sync(_client_error_steps(ListByteStream))
+
+
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_reconnect_does_not_retry_a_client_error():
+    with pytest.raises(NotFoundError):
+        await _run_async(_client_error_steps(AsyncListByteStream))
+
+
+# The first request isn't a reconnect, so it raises even when it names a Location.
+
+
+def _first_request_503(request):
+    return httpx.Response(
+        503,
+        headers={"Location": "/reconnect"},
+        json={"detail": "upstream"},
+        request=request,
+    )
+
+
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_first_request_failure_is_not_retried():
+    with pytest.raises(InternalServerError):
+        _run_sync([_first_request_503])
+
+
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_first_request_failure_is_not_retried():
+    with pytest.raises(InternalServerError):
+        await _run_async([_first_request_503])
+
+
+# A server that stays down gets five reconnects, backing off 1, 2, 4, 5 and 5s
+# (each plus up to 25% jitter), so about 20s, before the stream gives up. The
+# error it gives up with is caused by the last failure, so callers see why.
+
+GIVE_UP = "Exceeded maximum SSE reconnection attempts"
+
+
+def _server_down_steps(stream_cls):
+    return [
+        _ok(stream_cls, [_event(1)], drop=True, location="/reconnect"),
+        *[_status(502)] * 5,
+    ]
+
+
+def _assert_backed_off_for_about_20s(sleeps: list[float]) -> None:
+    assert len(sleeps) == 5
+    for slept, delay in zip(sleeps, [1, 2, 4, 5, 5], strict=True):
+        assert delay <= slept <= delay * 1.25
+
+
+def test_sync_reconnect_gives_up_after_five_backed_off_attempts(no_sleep):
+    with pytest.raises(httpx.TransportError, match=GIVE_UP) as exc_info:
+        _run_sync(_server_down_steps(ListByteStream))
+    assert isinstance(exc_info.value.__cause__, InternalServerError)
+    _assert_backed_off_for_about_20s(no_sleep)
+
+
+async def test_async_reconnect_gives_up_after_five_backed_off_attempts(no_sleep):
+    with pytest.raises(httpx.TransportError, match=GIVE_UP) as exc_info:
+        await _run_async(_server_down_steps(AsyncListByteStream))
+    assert isinstance(exc_info.value.__cause__, InternalServerError)
+    _assert_backed_off_for_about_20s(no_sleep)
+
+
+# A long stream that drops 6 times, each reconnect succeeding and making
+# progress, should not run out of attempts.
+
+
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_long_stream_survives_many_spaced_out_drops():
+    parts, _ = _run_sync(_long_stream_steps(ListByteStream, drops=6))
+    assert [p.data for p in parts] == [0, 1, 2, 3, 4, 5, 6, None]
+
+
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_long_stream_survives_many_spaced_out_drops():
+    parts, _ = await _run_async(_long_stream_steps(AsyncListByteStream, drops=6))
+    assert [p.data for p in parts] == [0, 1, 2, 3, 4, 5, 6, None]
+
+
+# Each reconnect resumes after the newest event the stream got, so a reconnect
+# loop on a resumable run replays only what it missed. A failed reconnect keeps
+# the cursor for the next one.
+RECONNECT_LOOPS = {
+    "progress_between_drops": (
+        lambda stream_cls: _long_stream_steps(stream_cls, drops=3),
+        [None, "0", "1", "2"],
+    ),
+    "failed_reconnect": (
+        lambda stream_cls: _transient_steps(stream_cls, _status(502)),
+        [None, "1", "1"],
+    ),
+}
+
+
+@pytest.mark.usefixtures("no_sleep")
+@pytest.mark.parametrize(
+    ("steps", "cursors"), RECONNECT_LOOPS.values(), ids=RECONNECT_LOOPS.keys()
+)
+def test_sync_reconnect_loop_resumes_after_the_newest_event(steps, cursors):
+    _, seen = _run_sync(steps(ListByteStream))
+    assert [r.headers.get("last-event-id") for r in seen] == cursors
+
+
+@pytest.mark.usefixtures("no_sleep")
+@pytest.mark.parametrize(
+    ("steps", "cursors"), RECONNECT_LOOPS.values(), ids=RECONNECT_LOOPS.keys()
+)
+async def test_async_reconnect_loop_resumes_after_the_newest_event(steps, cursors):
+    _, seen = await _run_async(steps(AsyncListByteStream))
+    assert [r.headers.get("last-event-id") for r in seen] == cursors
+
+
+# Reconnects should back off rather than hammer the server.
+
+
+def test_sync_reconnect_backs_off(no_sleep):
+    _run_sync(_long_stream_steps(ListByteStream, drops=2))
+    assert len(no_sleep) == 2 and all(d > 0 for d in no_sleep)
+
+
+async def test_async_reconnect_backs_off(no_sleep):
+    await _run_async(_long_stream_steps(AsyncListByteStream, drops=2))
+    assert len(no_sleep) == 2 and all(d > 0 for d in no_sleep)
+
+
 # --- _sse_to_v2_dict conversion ---


@@ -0,0 +1,128 @@
+import dataclasses
+
+import pytest
+
+import langgraph_sdk
+from langgraph_sdk import Priority, PriorityContext, PrioritySettings
+from langgraph_sdk.encryption import DuplicateHandlerError, LangGraphBetaWarning
+from langgraph_sdk.priority import _warn_priority_beta
+
+
+class _User:
+    is_authenticated = True
+    display_name = "user-1"
+    identity = "user-1"
+    permissions: tuple[str, ...] = ()
+
+    def __getitem__(self, key):
+        return getattr(self, key)
+
+    def __contains__(self, key):
+        return hasattr(self, key)
+
+    def __iter__(self):
+        return iter(("identity", "display_name", "is_authenticated", "permissions"))
+
+
+def _context(thread_id: str | None) -> PriorityContext:
+    return PriorityContext(
+        user=_User(),
+        permissions=[],
+        thread_id=thread_id,
+        thread_metadata={},
+        assistant_id="assistant-1",
+        graph_id="agent",
+        assistant_metadata={"tier": "gold"},
+        run_metadata={},
+        cron_id=None,
+    )
+
+
+async def _no_params() -> PrioritySettings | None:
+    return None
+
+
+async def _two_params(_ctx: PriorityContext, _extra: object) -> PrioritySettings | None:
+    return None
+
+
+def test_priority_and_its_types_are_exported():
+    assert {"Priority", "PriorityContext", "PrioritySettings"} <= set(
+        langgraph_sdk.__all__
+    )
+    assert Priority.types.PriorityContext is PriorityContext
+    assert Priority.types.PrioritySettings is PrioritySettings
+
+
+def test_construction_warns_that_priority_is_beta():
+    # The warning fires once per process, so an earlier test may have used it up.
+    _warn_priority_beta.cache_clear()
+    with pytest.warns(LangGraphBetaWarning, match="Priority API is in beta"):
+        Priority()
+
+
+def test_assign_registers_and_returns_the_function():
+    priority = Priority()
+
+    async def assign(_ctx: PriorityContext) -> PrioritySettings | None:
+        return {"priority": 3, "fairness_group": "org-42"}
+
+    assert priority.assign(assign) is assign
+    assert priority._assign_handler is assign
+
+
+def test_assign_rejects_a_sync_function():
+    priority = Priority()
+
+    def assign(_ctx: PriorityContext) -> PrioritySettings | None:
+        return None
+
+    with pytest.raises(TypeError, match="must be an async function"):
+        priority.assign(assign)  # ty: ignore[invalid-argument-type]
+    assert priority._assign_handler is None
+
+
+@pytest.mark.parametrize("fn", [_no_params, _two_params])
+def test_assign_rejects_the_wrong_number_of_parameters(fn):
+    priority = Priority()
+
+    with pytest.raises(TypeError, match="must accept exactly 1 parameter"):
+        priority.assign(fn)
+    assert priority._assign_handler is None
+
+
+def test_assign_rejects_a_second_function():
+    priority = Priority()
+
+    @priority.assign
+    async def first(_ctx: PriorityContext) -> PrioritySettings | None:
+        return None
+
+    with pytest.raises(DuplicateHandlerError):
+
+        @priority.assign
+        async def second(_ctx: PriorityContext) -> PrioritySettings | None:
+            return None
+
+    assert priority._assign_handler is first
+
+
+def test_context_accepts_no_thread_id():
+    # `make lint` type-checks this call too, so `thread_id` must accept None there.
+    assert _context(thread_id=None).thread_id is None
+
+
+def test_context_is_frozen():
+    ctx = _context(thread_id="thread-1")
+
+    with pytest.raises(dataclasses.FrozenInstanceError):
+        ctx.thread_id = "thread-2"  # ty: ignore[invalid-assignment]
+
+
+def test_every_settings_key_is_optional():
+    assert PrioritySettings.__optional_keys__ == {
+        "priority",
+        "fairness_group",
+        "fairness_weight",
+    }
+    assert PrioritySettings.__required_keys__ == frozenset()
@@ -0,0 +1,214 @@
+from __future__ import annotations
+
+import json
+from collections.abc import AsyncIterator, Callable, Iterator, Sequence
+from typing import Any
+
+import httpx
+import pytest
+
+from langgraph_sdk.client import (
+    HttpClient,
+    RunsClient,
+    SyncHttpClient,
+    SyncRunsClient,
+)
+
+
+class _BodyDropsMidRead(httpx.SyncByteStream, httpx.AsyncByteStream):
+    def __iter__(self) -> Iterator[bytes]:
+        yield b'{"partial'
+        raise httpx.ReadError("connection dropped")
+
+    async def __aiter__(self) -> AsyncIterator[bytes]:
+        yield b'{"partial'
+        raise httpx.ReadError("connection dropped")
+
+
+def _done(_request: httpx.Request) -> httpx.Response:
+    return httpx.Response(200, json={"done": True})
+
+
+def _wait_handler(
+    requests: list[httpx.Request],
+    reconnects: Sequence[Callable[[httpx.Request], httpx.Response]] = (_done,),
+):
+    """POST drops mid-body; the reconnects answer from `reconnects` in order."""
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        requests.append(request)
+        if request.method == "POST":
+            return httpx.Response(
+                200,
+                headers={"Location": "/threads/t1/runs/r1/join"},
+                stream=_BodyDropsMidRead(),
+            )
+        return reconnects[len(requests) - 2](request)
+
+    return handler
+
+
+def _assert_bodyless_get(request: httpx.Request) -> None:
+    assert request.method == "GET"
+    assert request.url.path == "/threads/t1/runs/r1/join"
+    assert "content-length" not in request.headers, request.headers
+    assert "content-type" not in request.headers, request.headers
+    assert request.read() == b""
+
+
+@pytest.mark.usefixtures("no_sleep")
+def test_sync_wait_reconnects_with_a_bodyless_get():
+    requests: list[httpx.Request] = []
+    transport = httpx.MockTransport(_wait_handler(requests))
+    with httpx.Client(transport=transport, base_url="https://example.com") as client:
+        runs = SyncRunsClient(SyncHttpClient(client))
+        with pytest.warns(UserWarning, match="attempting reconnect"):
+            result = runs.wait("t1", "agent", input={"question": "hi"})
+
+    assert result == {"done": True}
+    _assert_bodyless_get(requests[1])
+
+
+@pytest.mark.usefixtures("no_sleep")
+async def test_async_wait_reconnects_with_a_bodyless_get():
+    requests: list[httpx.Request] = []
+    transport = httpx.MockTransport(_wait_handler(requests))
+    async with httpx.AsyncClient(
+        transport=transport, base_url="https://example.com"
+    ) as client:
+        runs = RunsClient(HttpClient(client))
+        with pytest.warns(UserWarning, match="attempting reconnect"):
+            result = await runs.wait("t1", "agent", input={"question": "hi"})
+
+    assert result == {"done": True}
+    _assert_bodyless_get(requests[1])
+
+
+def _connect_error(request: httpx.Request) -> httpx.Response:
+    raise httpx.ConnectError("[Errno 61] Connection refused", request=request)
+
+
+async def _wait(
+    handler: Callable[[httpx.Request], httpx.Response], *, asynchronous: bool
+) -> Any:
+    transport = httpx.MockTransport(handler)
+    if asynchronous:
+        async with httpx.AsyncClient(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            return await RunsClient(HttpClient(client)).wait("t1", "agent", input={})
+    with httpx.Client(transport=transport, base_url="https://example.com") as client:
+        return SyncRunsClient(SyncHttpClient(client)).wait("t1", "agent", input={})
+
+
+# What a server restart or a deploy looks like to the reconnect GET: each one is
+# retried after a backoff.
+RECONNECT_FAILURES = {
+    "429": lambda request: httpx.Response(429, request=request),
+    "503": lambda request: httpx.Response(503, request=request),
+    "connect_error": _connect_error,
+    "body_drops": lambda _request: httpx.Response(
+        200,
+        headers={"Location": "/threads/t1/runs/r1/join"},
+        stream=_BodyDropsMidRead(),
+    ),
+}
+
+
+@pytest.mark.parametrize("asynchronous", [False, True])
+@pytest.mark.parametrize(
+    "failure", RECONNECT_FAILURES.values(), ids=RECONNECT_FAILURES.keys()
+)
+async def test_wait_retries_a_failed_reconnect_after_a_backoff(
+    failure, asynchronous, no_sleep
+):
+    requests: list[httpx.Request] = []
+    with pytest.warns(UserWarning, match="attempting reconnect"):
+        result = await _wait(
+            _wait_handler(requests, [failure, _done]), asynchronous=asynchronous
+        )
+
+    assert result == {"done": True}
+    assert len(requests) == 3
+    assert len(no_sleep) == 2 and min(no_sleep) >= 1, no_sleep
+
+
+@pytest.mark.usefixtures("no_sleep")
+@pytest.mark.parametrize("asynchronous", [False, True])
+async def test_wait_raises_after_five_failed_reconnects(asynchronous):
+    requests: list[httpx.Request] = []
+    with (
+        pytest.warns(UserWarning, match="attempting reconnect"),
+        pytest.raises(httpx.HTTPStatusError, match="503"),
+    ):
+        await _wait(
+            _wait_handler(requests, [RECONNECT_FAILURES["503"]] * 5),
+            asynchronous=asynchronous,
+        )
+
+    assert len(requests) == 6
+
+
+@pytest.mark.usefixtures("no_sleep")
+@pytest.mark.parametrize("asynchronous", [False, True])
+async def test_wait_does_not_retry_a_client_error(asynchronous):
+    requests: list[httpx.Request] = []
+    with (
+        pytest.warns(UserWarning, match="attempting reconnect"),
+        pytest.raises(httpx.HTTPStatusError, match="404"),
+    ):
+        await _wait(
+            _wait_handler(
+                requests, [lambda request: httpx.Response(404, request=request)]
+            ),
+            asynchronous=asynchronous,
+        )
+
+    assert len(requests) == 2
+
+
+def test_sync_wait_raises_the_run_error():
+    requests: list[httpx.Request] = []
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        requests.append(request)
+        return httpx.Response(
+            200, json={"__error__": {"error": "RuntimeError", "message": "boom"}}
+        )
+
+    transport = httpx.MockTransport(handler)
+    with httpx.Client(transport=transport, base_url="https://example.com") as client:
+        runs = SyncRunsClient(SyncHttpClient(client))
+        with pytest.raises(Exception, match="RuntimeError: boom"):
+            runs.wait("t1", "agent", input={}, raise_error=True)
+
+    assert "raise_error" not in json.loads(requests[0].content)
+
+
+@pytest.mark.parametrize("asynchronous", [False, True])
+async def test_wait_returns_the_run_error_without_raise_error(asynchronous):
+    requests: list[httpx.Request] = []
+    run_error = {"__error__": {"error": "RuntimeError", "message": "boom"}}
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        requests.append(request)
+        return httpx.Response(200, json=run_error)
+
+    transport = httpx.MockTransport(handler)
+    if asynchronous:
+        async with httpx.AsyncClient(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            result = await RunsClient(HttpClient(client)).wait(
+                "t1", "agent", input={}, raise_error=False
+            )
+    else:
+        with httpx.Client(
+            transport=transport, base_url="https://example.com"
+        ) as client:
+            result = SyncRunsClient(SyncHttpClient(client)).wait(
+                "t1", "agent", input={}, raise_error=False
+            )
+
+    assert result == run_error
+    assert "raise_error" not in json.loads(requests[0].content)
@@ -1,5 +1,5 @@
 version = 1
-revision = 5
+revision = 3
 requires-python = ">=3.10"
 resolution-markers = [
    "python_full_version >= '3.15'",
@@ -40,11 +40,11 @@ wheels = [

 [[package]]
 name = "certifi"
-version = "2026.2.25"
+version = "2026.10.19"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/af/2d/7bf41579a8986e348fa033a31cdd0e4121114f6bce2457e8876010b092dd/certifi-2026.2.25.tar.gz", hash = "sha256:e887ab5cee78ea814d3472169153c2d12cd43b14bd03329a39a9c6e2e80bfba7", size = 155029, upload-time = "2026-02-25T02:54:17.342Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/a6/bd/1436c50cc46699de0230fbaacc1d1de7bfec413176099949e9d5112e88f2/certifi-2026.10.19.tar.gz", hash = "sha256:035c1e85572b141bf1c3a3eff0e33a8eb74917522f2559075ff8d2fea780a1ae", size = 137598, upload-time = "2026-10-20T02:24:16.604Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/9a/3c/c17fb3ca2d9c3acff52e30b309f538586f9f5b9c9cf454f3845fc9af4881/certifi-2026.2.25-py3-none-any.whl", hash = "sha256:027692e4402ad994f1c42e52a4997a9763c646b73e4096e4d5d6db8af1d6f0fa", size = 153684, upload-time = "2026-02-25T02:54:15.766Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/51/29046e48f693c54e5b7412216d447a1d8362c9653148d98c02cdf65fa733/certifi-2026.10.19-py3-none-any.whl", hash = "sha256:42d35340cc5b9ba78e818ddbff0c5b1c2b779de448a340fd1b989f5fa8670e1b", size = 136464, upload-time = "2026-10-20T02:24:15.072Z" },
 ]

 [[package]]
@@ -301,7 +301,7 @@ wheels = [

 [[package]]
 name = "langgraph"
-version = "1.2.16"
+version = "1.2.19"
 source = { editable = "../langgraph" }
 dependencies = [
    { name = "langchain-core" },
@@ -385,7 +385,7 @@ test = [

 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
    { name = "langchain-core" },
@@ -525,7 +525,7 @@ requires-dist = [
    { name = "langchain-core", specifier = ">=1.4.0,<2" },
    { name = "langchain-protocol", specifier = ">=0.0.15" },
    { name = "orjson", specifier = ">=3.11.5" },
-    { name = "websockets", specifier = ">=14,<17" },
+    { name = "websockets", specifier = ">=14" },
 ]

 [package.metadata.requires-dev]