fix(langgraph): don't report a paused tool as failed on the tools stream (#9406)

`interrupt()` inside a tool raises `GraphInterrupt`, and langchain-core
calls `on_tool_error` for every exception, so the `tools` stream
reported a tool that was only waiting for an answer as `tool-error`,
with the interrupt flattened into the message. With a subagent running
inside a tool, the outer tool showed as failed while the subagent waited
for approval. Fixes #8218.

## Fix

- An interrupt or a drain ends the tool's events without a `tool-error`.
On resume the tool runs again and emits `tool-started` and
`tool-finished`. A parent command doesn't rerun the tool, so it still
ends the call with `tool-error`, as before.
- `astream` no longer runs every node in streaming mode just because the
`tools` handler is attached. Like the `messages` handler, it doesn't
need it, and with it a `@task` awaited inside a subgraph node returned
`None` whenever the `tools` mode was on.

The new tests fail on `main`.

## Release note

The `tools` stream no longer sends `tool-error` for a tool that pauses
on `interrupt()`. Until the run resumes, that tool has a `tool-started`
event and no end event. In-process `stream_events(version="v3")` now
ends that tool's stream as completed with no output, instead of failed
with the interrupt.

Same idea as #8219 by @haseeb-m4235 and #8282 by @thsfranca.

AGENTS.md

@@ -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.

libs/checkpoint-conformance/uv.lock

@@ -279,7 +279,7 @@ wheels = [
 
 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
     { name = "langchain-core" },

libs/checkpoint-postgres/uv.lock

@@ -278,7 +278,7 @@ wheels = [
 
 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
     { name = "langchain-core" },

libs/checkpoint-sqlite/uv.lock

@@ -285,7 +285,7 @@ wheels = [
 
 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
     { name = "langchain-core" },

libs/checkpoint/pyproject.toml

@@ -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"

libs/checkpoint/uv.lock

@@ -301,7 +301,7 @@ wheels = [
 
 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "." }
 dependencies = [
     { name = "langchain-core" },

libs/cli/langgraph_cli/__init__.py

@@ -1 +1 @@
-__version__ = "0.4.38"
+__version__ = "0.4.40"

libs/cli/langgraph_cli/config.py

@@ -1738,7 +1738,6 @@ def docker_tag(
         api_version = _resolve_compatible_api_version(
             api_version, base_image, f"{language}{version_distro_tag}"
         )
-        api_version = re.sub(r"(?<=\d)dev(?=\d+$)", ".dev", api_version)
         full_tag = f"{api_version}-{language}{version_distro_tag}"
     elif "/langgraph-server" in base_image and version_distro_tag not in base_image:
         return f"{base_image}-{language}{version_distro_tag}"

libs/cli/langgraph_cli/deploy.py

@@ -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)))
 
 
 # ---------------------------------------------------------------------------

libs/cli/langgraph_cli/host_backend.py

@@ -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(

libs/cli/tests/unit_tests/cli/test_cli.py

@@ -482,6 +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] = {}
@@ -564,6 +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()
 
@@ -766,6 +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] = {}

libs/cli/tests/unit_tests/cli/test_deploy_command.py

@@ -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(

libs/cli/tests/unit_tests/test_config.py

@@ -3333,37 +3333,20 @@ def test_docker_tag_with_api_version(in_config: bool):
 
 
 @pytest.mark.parametrize("in_config", [False, True])
-@pytest.mark.parametrize(
-    "runtime,image", [("python", "langgraph-api"), ("node", "langgraphjs-api")]
-)
-@pytest.mark.parametrize(
-    "version,expected",
-    [
-        ("0.9.0rc1", "0.9.0rc1"),
-        ("0.9.0.dev1", "0.9.0.dev1"),
-        ("0.18.0dev11", "0.18.0.dev11"),
-    ],
-)
-def test_docker_tag_with_prerelease_api_version(
-    version: str, expected: str, in_config: bool, runtime: str, image: str
-):
+@pytest.mark.parametrize("version", ["0.9.0rc1", "0.9.0.dev1"])
+def test_docker_tag_with_prerelease_api_version(version: str, in_config: bool):
     """Test docker_tag with prerelease and dev api_version values."""
     config = validate_config(
         {
-            f"{runtime}_version": "3.12" if runtime == "python" else "20",
+            "python_version": "3.11",
             "dependencies": ["."],
-            "graphs": {
-                "agent": "./agent.py:graph"
-                if runtime == "python"
-                else "./agent.js:graph"
-            },
+            "graphs": {"agent": "./agent.py:graph"},
             "api_version": version if in_config else None,
         }
     )
 
     tag = docker_tag(config, api_version=version if not in_config else None)
-    suffix = "py3.12" if runtime == "python" else "node20"
-    assert tag == f"langchain/{image}:{expected}-{suffix}"
+    assert tag == f"langchain/langgraph-api:{version}-py3.11"
 
 
 def test_docker_tag_with_compatible_api_version_promotes_to_latest_patch():
@@ -3395,8 +3378,7 @@ def test_docker_tag_with_compatible_api_version_promotes_to_latest_patch():
     assert tag == "langchain/langgraph-api:0.11.1-py3.12-wolfi"
 
 
-@pytest.mark.parametrize("dev_version", ["0.11.0.dev5", "0.11.0dev5"])
-def test_docker_tag_with_compatible_api_version_freezes_dev_until_rc(dev_version: str):
+def test_docker_tag_with_compatible_api_version_freezes_dev_until_rc():
     config = validate_config(
         {
             "python_version": "3.12",
@@ -3409,7 +3391,7 @@ def test_docker_tag_with_compatible_api_version_freezes_dev_until_rc(dev_version
     with patch(
         "langgraph_cli.config._get_pypi_versions",
         return_value=[
-            dev_version,
+            "0.11.0.dev5",
             "0.11.0.dev6",
             "0.11.0.dev7",
         ],

libs/cli/tests/unit_tests/test_deploy_agents.py

@@ -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"] == {

libs/cli/tests/unit_tests/test_deploy_helpers.py

@@ -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

libs/cli/tests/unit_tests/test_host_backend.py

@@ -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"),
     [

libs/cli/tests/unit_tests/test_util.py

@@ -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", "----  ------"]

libs/langgraph/langgraph/_internal/_config.py

@@ -347,6 +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(
             {
@@ -355,6 +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
@@ -407,12 +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]

libs/langgraph/langgraph/channels/delta.py

@@ -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(

libs/langgraph/langgraph/pregel/_draw.py

@@ -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:

libs/langgraph/langgraph/pregel/_executor.py

@@ -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():

libs/langgraph/langgraph/pregel/_loop.py

@@ -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
 

libs/langgraph/langgraph/pregel/_retry.py

@@ -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):

libs/langgraph/langgraph/pregel/_runner.py

@@ -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)

libs/langgraph/langgraph/pregel/_tools.py

@@ -9,6 +9,7 @@ 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.protocol import StreamChunk
 
 try:
@@ -188,6 +189,10 @@ 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,

libs/langgraph/langgraph/pregel/main.py

@@ -3310,7 +3310,9 @@ class Pregel(
                         True
                         for h in run_manager.handlers
                         if isinstance(h, _StreamingCallbackHandler)
-                        and not isinstance(h, StreamMessagesHandler)
+                        and not isinstance(
+                            h, (StreamMessagesHandler, StreamToolCallHandler)
+                        )
                     ),
                     False,
                 )

libs/langgraph/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "langgraph"
-version = "1.2.17"
+version = "1.2.18"
 description = "Building stateful, multi-actor applications with LLMs"
 authors = []
 requires-python = ">=3.10"

libs/langgraph/tests/__snapshots__/test_large_cases.ambr

@@ -133,6 +133,11 @@
             "ToolNode"
           ],
           "name": "tools"
+        },
+        "metadata": {
+          "__tools": [
+            "search_api"
+          ]
         }
       },
       {
@@ -221,6 +226,11 @@
             "ToolNode"
           ],
           "name": "tools"
+        },
+        "metadata": {
+          "__tools": [
+            "search_api"
+          ]
         }
       },
       {

libs/langgraph/tests/test_channels.py

@@ -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.
 

libs/langgraph/tests/test_pregel.py

@@ -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 = (

libs/langgraph/tests/test_pregel_async.py

@@ -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:

libs/langgraph/tests/test_retry.py

@@ -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

libs/langgraph/tests/test_tool_stream_handler.py

@@ -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:

libs/langgraph/tests/test_utils.py

@@ -799,6 +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")

libs/langgraph/uv.lock

@@ -1439,7 +1439,7 @@ wheels = [
 
 [[package]]
 name = "langgraph"
-version = "1.2.17"
+version = "1.2.18"
 source = { editable = "." }
 dependencies = [
     { name = "langchain-core" },
@@ -1625,7 +1625,7 @@ wheels = [
 
 [[package]]
 name = "langgraph-checkpoint"
-version = "4.3.0"
+version = "4.3.1"
 source = { editable = "../checkpoint" }
 dependencies = [
     { name = "langchain-core" },

libs/prebuilt/uv.lock

@@ -286,7 +286,7 @@ wheels = [
 
 [[package]]
 name = "langgraph"
-version = "1.2.17"
+version = "1.2.18"
 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" },

libs/sdk-py/langgraph_sdk/__init__.py

@@ -5,7 +5,7 @@ from langgraph_sdk.encryption.types import DecryptResult, EncryptionContext
 from langgraph_sdk.priority import Priority
 from langgraph_sdk.priority.types import PriorityContext, PrioritySettings
 
-__version__ = "0.4.9"
+__version__ = "0.4.10"
 
 __all__ = [
     "Auth",

libs/sdk-py/langgraph_sdk/_async/http.py

@@ -179,7 +179,7 @@ class HttpClient:
                 return await self.request_reconnect(
                     loc,
                     "GET",
-                    headers=request_headers,
+                    headers=headers,
                     # don't pass on_response so it's only called once
                     reconnect_limit=reconnect_limit - 1,
                 )
@@ -218,6 +218,12 @@ 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
 
             current_method = method if reconnect_path is None else "GET"

libs/sdk-py/langgraph_sdk/_async/runs.py

@@ -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(

libs/sdk-py/langgraph_sdk/_sync/http.py

@@ -179,7 +179,7 @@ class SyncHttpClient:
                 return self.request_reconnect(
                     loc,
                     "GET",
-                    headers=request_headers,
+                    headers=headers,
                     # don't pass on_response so it's only called once
                     reconnect_limit=reconnect_limit - 1,
                 )
@@ -220,6 +220,12 @@ 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
 
             current_method = method if reconnect_path is None else "GET"
@@ -286,7 +292,9 @@ class SyncHttpClient:
             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,

libs/sdk-py/langgraph_sdk/_sync/runs.py

@@ -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,

libs/sdk-py/langgraph_sdk/sse.py

@@ -79,23 +79,32 @@ class SSEDecoder:
     def __init__(self) -> None:
         self._event = ""
         self._data = bytearray()
-        self._last_event_id = ""
+        self._last_event_id: str | None = None
         self._retry: int | None = None
+        self._at_stream_start = True
 
     @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 the stream sends an `id` field, 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:
             if (
                 not self._event
                 and not self._data
-                and not self._last_event_id
+                and self._last_event_id is None
                 and self._retry is None
             ):
                 return None
@@ -103,7 +112,7 @@ class SSEDecoder:
             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,
+                id=self._last_event_id or None,
             )
 
             # NOTE: as per the SSE spec, do not reset last_event_id.
@@ -124,6 +133,8 @@ class SSEDecoder:
         if fieldname == b"event":
             self._event = value.decode()
         elif fieldname == b"data":
+            if self._data:
+                self._data.extend(b"\n")
             self._data.extend(value)
         elif fieldname == b"id":
             if b"\0" in value:

libs/sdk-py/tests/test_api_parity.py

@@ -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

libs/sdk-py/tests/test_client_stream.py

@@ -5,6 +5,7 @@ from pathlib import Path
 from typing import Any
 
 import httpx
+import orjson
 import pytest
 from typing_extensions import assert_type
 
@@ -110,6 +111,25 @@ 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]
+
+
 # --- HTTP client streaming ---
 
 
@@ -292,6 +312,121 @@ 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
+def test_sync_stream_reconnects_without_a_cursor_after_an_empty_id(
+    empty_id: bytes, headers: dict[str, str] | None
+):
+    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
+async def test_async_stream_reconnects_without_a_cursor_after_an_empty_id(
+    empty_id: bytes, headers: dict[str, str] | None
+):
+    reconnect = await _async_reconnect(
+        [b"id: 1\nevent: values\ndata: 1\n\n", empty_id], headers
+    )
+
+    assert "last-event-id" not in reconnect.headers
+
+
+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"]
+
+
+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"]
+
+
+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"]
+
+
+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"]
+
+
 # --- _sse_to_v2_dict conversion ---
 
 

libs/sdk-py/tests/test_runs_client.py

@@ -0,0 +1,119 @@
+from __future__ import annotations
+
+import json
+from collections.abc import AsyncIterator, Iterator
+
+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 _wait_handler(requests: list[httpx.Request]):
+    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 httpx.Response(200, json={"done": True})
+
+    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""
+
+
+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])
+
+
+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 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)

libs/sdk-py/uv.lock

@@ -301,7 +301,7 @@ wheels = [
 
 [[package]]
 name = "langgraph"
-version = "1.2.17"
+version = "1.2.18"
 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" },