One more

minor fix

.github/workflows/_test_release.yml

@@ -13,7 +13,7 @@ env:
 
 jobs:
   build:
-    if: github.ref == 'refs/heads/main'
+    if: github.ref == 'refs/heads/main' || github.ref == 'refs/heads/v0'
     runs-on: ubuntu-latest
 
     outputs:

.github/workflows/release.yml

@@ -13,7 +13,7 @@ env:
 
 jobs:
   build:
-    if: github.ref == 'refs/heads/main'
+    if: github.ref == 'refs/heads/main' || github.ref == 'refs/heads/v0'
     runs-on: ubuntu-latest
 
     outputs:

README.md

@@ -74,7 +74,7 @@ While LangGraph can be used standalone, it also integrates seamlessly with any L
 
 - [Guides](https://langchain-ai.github.io/langgraph/how-tos/): Quick, actionable code snippets for topics such as streaming, adding memory & persistence, and design patterns (e.g. branching, subgraphs, etc.).
 - [Reference](https://langchain-ai.github.io/langgraph/reference/graphs/): Detailed reference on core classes, methods, how to use the graph and checkpointing APIs, and higher-level prebuilt components.
-- [Examples](https://langchain-ai.github.io/langgraph/tutorials/): Guided examples on getting started with LangGraph.
+- [Examples](https://langchain-ai.github.io/langgraph/tutorials/overview/): Guided examples on getting started with LangGraph.
 - [LangChain Academy](https://academy.langchain.com/courses/intro-to-langgraph): Learn the basics of LangGraph in our free, structured course.
 - [Templates](https://langchain-ai.github.io/langgraph/concepts/template_applications/): Pre-built reference apps for common agentic workflows (e.g. ReAct agent, memory, retrieval etc.) that can be cloned and adapted.
 - [Case studies](https://www.langchain.com/built-with-langgraph): Hear how industry leaders use LangGraph to ship AI applications at scale.

docs/docs/agents/agents.md

@@ -15,7 +15,7 @@ This guide shows you how to set up and use LangGraph's **prebuilt**, **reusable*
 
 Before you start this tutorial, ensure you have the following:
 
-- An [Anthropic](https://console.anthropic.com/settings/admin-keys) API key 
+- An [Anthropic](https://console.anthropic.com/settings/keys) API key 
 
 ## 1. Install dependencies
 

docs/docs/agents/memory.md

@@ -82,7 +82,7 @@ ny_response = agent.invoke(
 ```
 
 1. The `InMemorySaver` is a checkpointer that stores the agent's state in memory. In a production setting, you would typically use a database or other persistent storage. Please review the [checkpointer documentation](../reference/checkpoints.md) for more options. If you're deploying with **LangGraph Platform**, the platform will provide a production-ready checkpointer for you.
-2. The `checkpointer` is passed to the agent. This enables the agent to persist its state across invocations. Please note that 
+2. The `checkpointer` is passed to the agent. This enables the agent to persist its state across invocations.
 3. A unique `thread_id` is provided in the config. This ID is used to identify the conversation session. The value is controlled by the user and can be any string.
 4. The agent will continue the conversation using the same `thread_id`. This will allow the agent to infer that the user is asking specifically about the **weather** in New York. 
 

docs/docs/agents/overview.md

@@ -53,3 +53,139 @@ The high-level components are organized into several packages, each with a speci
 | `langmem`                                  | Agent memory management: [**short-term and long-term**](./memory.md)        | `pip install -U langmem`                |
 | `agentevals`                               | Utilities to [**evaluate agent performance**](./evals.md)                   | `pip install -U agentevals`             |
 
+## Visualize an agent graph
+
+Use the following tool to visualize the graph generated by
+[`create_react_agent`][langgraph.prebuilt.chat_agent_executor.create_react_agent]
+and to view an outline of the corresponding code.
+It allows you to explore the infrastructure of the agent as defined by the presence of:
+
+* [`tools`](../agents/tools.md): A list of tools (functions, APIs, or other callable objects) that the agent can use to perform tasks.
+* [`pre_model_hook`](../how-tos/create-react-agent-manage-message-history.ipynb): A function that is called before the model is invoked. It can be used to condense messages or perform other preprocessing tasks.
+* `post_model_hook`: A function that is called after the model is invoked. It can be used to implement guardrails, human-in-the-loop flows, or other postprocessing tasks.
+* [`response_format`](../agents/agents.md#6-configure-structured-output): A data structure used to constrain the type of the final output, e.g., a `pydantic` `BaseModel`.
+
+<div class="agent-layout">
+  <div class="agent-graph-features-container">
+    <div class="agent-graph-features">
+      <h3 class="agent-section-title">Features</h3>
+      <label><input type="checkbox" id="tools" checked> <code>tools</code></label>
+      <label><input type="checkbox" id="pre_model_hook"> <code>pre_model_hook</code></label>
+      <label><input type="checkbox" id="post_model_hook"> <code>post_model_hook</code></label>
+      <label><input type="checkbox" id="response_format"> <code>response_format</code></label>
+    </div>
+  </div>
+
+  <div class="agent-graph-container">
+    <h3 class="agent-section-title">Graph</h3>
+    <img id="agent-graph-img" src="../assets/react_agent_graphs/0001.svg" alt="graph image" style="max-width: 100%;"/>
+  </div>
+</div>
+
+
+The following code snippet shows how to create the above agent (and underlying graph) with
+[`create_react_agent`][langgraph.prebuilt.chat_agent_executor.create_react_agent]:
+
+<div class="language-python">
+  <pre><code id="agent-code" class="language-python"></code></pre>
+</div>
+
+
+<script>
+function getCheckedValue(id) {
+  return document.getElementById(id).checked ? "1" : "0";
+}
+
+function getKey() {
+  return [
+    getCheckedValue("response_format"),
+    getCheckedValue("post_model_hook"),
+    getCheckedValue("pre_model_hook"),
+    getCheckedValue("tools")
+  ].join("");
+}
+
+function generateCodeSnippet({ tools, pre, post, response }) {
+  const lines = [
+    "from langgraph.prebuilt import create_react_agent",
+    "from langchain_openai import ChatOpenAI"
+  ];
+
+  if (response) lines.push("from pydantic import BaseModel");
+
+  lines.push("", 'model = ChatOpenAI("o4-mini")', "");
+
+  if (tools) {
+    lines.push(
+      "def tool() -> None:",
+      '    """Testing tool."""',
+      "    ...",
+      ""
+    );
+  }
+
+  if (pre) {
+    lines.push(
+      "def pre_model_hook() -> None:",
+      '    """Pre-model hook."""',
+      "    ...",
+      ""
+    );
+  }
+
+  if (post) {
+    lines.push(
+      "def post_model_hook() -> None:",
+      '    """Post-model hook."""',
+      "    ...",
+      ""
+    );
+  }
+
+  if (response) {
+    lines.push(
+      "class ResponseFormat(BaseModel):",
+      '    """Response format for the agent."""',
+      "    result: str",
+      ""
+    );
+  }
+
+  lines.push("agent = create_react_agent(");
+  lines.push("    model,");
+
+  if (tools) lines.push("    tools=[tool],");
+  if (pre) lines.push("    pre_model_hook=pre_model_hook,");
+  if (post) lines.push("    post_model_hook=post_model_hook,");
+  if (response) lines.push("    response_format=ResponseFormat,");
+
+  lines.push(")", "", "agent.get_graph().draw_mermaid_png()");
+
+  return lines.join("\n");
+}
+
+async function render() {
+  const key = getKey();
+  document.getElementById("agent-graph-img").src = `../assets/react_agent_graphs/${key}.svg`;
+
+  const state = {
+    tools: document.getElementById("tools").checked,
+    pre: document.getElementById("pre_model_hook").checked,
+    post: document.getElementById("post_model_hook").checked,
+    response: document.getElementById("response_format").checked
+  };
+
+  document.getElementById("agent-code").textContent = generateCodeSnippet(state);
+}
+
+function initializeWidget() {
+  render(); // no need for `await` here
+  document.querySelectorAll(".agent-graph-features input").forEach((input) => {
+    input.addEventListener("change", render);
+  });
+}
+
+// Init for both full reload and SPA nav (used by MkDocs Material)
+window.addEventListener("DOMContentLoaded", initializeWidget);
+document$.subscribe(initializeWidget);
+</script>

docs/docs/agents/tools.md

@@ -280,7 +280,21 @@ LangGraph allows access to short-term and long-term memory from tools. See [Memo
 
 ## Prebuilt tools
 
-LangChain supports a wide range of prebuilt tool integrations for interacting with APIs, databases, file systems, web data, and more. These tools extend the functionality of agents and enable rapid development.
+You can use prebuilt tools from model providers by passing a dictionary with tool specs to the `tools` parameter of `create_react_agent`. For example, to use the `web_search_preview` tool from OpenAI:
+
+```python
+from langgraph.prebuilt import create_react_agent
+
+agent = create_react_agent(
+    model="openai:gpt-4o-mini", 
+    tools=[{"type": "web_search_preview"}]
+)
+response = agent.invoke(
+    {"messages": ["What was a positive news story from today?"]}
+)
+```
+
+Additionally, LangChain supports a wide range of prebuilt tool integrations for interacting with APIs, databases, file systems, web data, and more. These tools extend the functionality of agents and enable rapid development.
 
 You can browse the full list of available integrations in the [LangChain integrations directory](https://python.langchain.com/docs/integrations/tools/).
 

docs/docs/assets/react-agent-graphs.json

@@ -0,0 +1,19 @@
+{
+    "0000": "graph TD;\n\t__start__ --> agent;\n\tagent --> __end__;",
+    "0001": "graph TD;\n\t__start__ --> agent;\n\tagent -.-> __end__;\n\tagent -.-> tools;\n\ttools --> agent;",
+    "0010": "graph TD;\n\t__start__ --> pre_model_hook;\n\tpre_model_hook --> agent;\n\tagent --> __end__;",
+    "0011": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent -.-> __end__;\n\tagent -.-> tools;\n\tpre_model_hook --> agent;\n\ttools --> pre_model_hook;",
+    "0100": "graph TD;\n\t__start__ --> agent;\n\tagent --> post_model_hook;\n\tpost_model_hook --> __end__;",
+    "0101": "graph TD;\n\t__start__ --> agent;\n\tagent --> post_model_hook;\n\tpost_model_hook -.-> __end__;\n\tpost_model_hook -.-> agent;\n\tpost_model_hook -.-> tools;\n\ttools --> agent;",
+    "0110": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent --> post_model_hook;\n\tpre_model_hook --> agent;\n\tpost_model_hook --> __end__;",
+    "0111": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent --> post_model_hook;\n\tpost_model_hook -.-> __end__;\n\tpost_model_hook -.-> pre_model_hook;\n\tpost_model_hook -.-> tools;\n\tpre_model_hook --> agent;\n\ttools --> pre_model_hook;",
+    "1000": "graph TD;\n\t__start__ --> agent;\n\tagent --> generate_structured_response;\n\tgenerate_structured_response --> __end__;",
+    "1001": "graph TD;\n\t__start__ --> agent;\n\tagent -.-> generate_structured_response;\n\tagent -.-> tools;\n\ttools --> agent;\n\tgenerate_structured_response --> __end__;",
+    "1010": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent --> generate_structured_response;\n\tpre_model_hook --> agent;\n\tgenerate_structured_response --> __end__;",
+    "1011": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent -.-> generate_structured_response;\n\tagent -.-> tools;\n\tpre_model_hook --> agent;\n\ttools --> pre_model_hook;\n\tgenerate_structured_response --> __end__;",
+    "1100": "graph TD;\n\t__start__ --> agent;\n\tagent --> post_model_hook;\n\tpost_model_hook --> generate_structured_response;\n\tgenerate_structured_response --> __end__;",
+    "1101": "graph TD;\n\t__start__ --> agent;\n\tagent --> post_model_hook;\n\tpost_model_hook -.-> agent;\n\tpost_model_hook -.-> generate_structured_response;\n\tpost_model_hook -.-> tools;\n\ttools --> agent;\n\tgenerate_structured_response --> __end__;",
+    "1110": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent --> post_model_hook;\n\tpost_model_hook --> generate_structured_response;\n\tpre_model_hook --> agent;\n\tgenerate_structured_response --> __end__;",
+    "1111": "graph TD;\n\t__start__ --> pre_model_hook;\n\tagent --> post_model_hook;\n\tpost_model_hook -.-> generate_structured_response;\n\tpost_model_hook -.-> pre_model_hook;\n\tpost_model_hook -.-> tools;\n\tpre_model_hook --> agent;\n\ttools --> pre_model_hook;\n\tgenerate_structured_response --> __end__;"
+}
+  

docs/docs/cloud/deployment/self_hosted_control_plane.md

@@ -2,8 +2,8 @@
 
 Before deploying, review the [conceptual guide for the Self-Hosted Control Plane](../../concepts/langgraph_self_hosted_control_plane.md) deployment option.
 
-!!! important "Beta"
-    The Self-Hosted Control Plane deployment option is currently in beta stage.
+!!! info "Important"
+    The Self-Hosted Control Plane deployment option is currently in beta stage and requires an [Enterprise](../../concepts/plans.md) plan.
 
 ## Prerequisites
 

docs/docs/cloud/deployment/self_hosted_data_plane.md

@@ -2,8 +2,8 @@
 
 Before deploying, review the [conceptual guide for the Self-Hosted Data Plane](../../concepts/langgraph_self_hosted_data_plane.md) deployment option.
 
-!!! important "Beta"
-    The Self-Hosted Data Plane deployment option is currently in beta stage.
+!!! info "Important"
+    The Self-Hosted Data Plane deployment option is currently in beta stage and requires an [Enterprise](../../concepts/plans.md) plan.
 
 ## Prerequisites
 

docs/docs/concepts/deployment_options.md

@@ -40,8 +40,8 @@ For more information, please see:
 
 ## Self-Hosted Data Plane
 
-!!! important "Beta"
-    The Self-Hosted Data Plane deployment option is currently in beta stage.
+!!! info "Important"
+    The Self-Hosted Data Plane deployment option is currently in beta stage and requires an [Enterprise](../concepts/plans.md) plan.
 
 The [Self-Hosted Data Plane](./langgraph_self_hosted_data_plane.md) deployment option is a "hybrid" model for deployment where we manage the [control plane](./langgraph_control_plane.md) in our cloud and you manage the [data plane](./langgraph_data_plane.md) in your cloud. This option provides a way to securely manage your data plane infrastructure, while offloading control plane management to us.
 
@@ -56,8 +56,8 @@ For more information, please see:
 
 ## Self-Hosted Control Plane
 
-!!! important "Beta"
-    The Self-Hosted Control Plane deployment option is currently in beta stage.
+!!! info "Important"
+    The Self-Hosted Control Plane deployment option is currently in beta stage and requires an [Enterprise](../concepts/plans.md) plan.
 
 The [Self-Hosted Control Plane](./langgraph_self_hosted_control_plane.md) deployment option is a fully self-hosted model for deployment where you manage the [control plane](./langgraph_control_plane.md) and [data plane](./langgraph_data_plane.md) in your cloud. This option give you full control and responsibility of the control plane and data plane infrastructure.
 

docs/docs/concepts/langgraph_control_plane.md

@@ -47,17 +47,22 @@ This section describes various features of the control plane.
 
 For simplicity, the control plane offers two deployment types with different resource allocations: `Development` and `Production`.
 
-| **Deployment Type** | **CPU** | **Memory** | **Scaling**         |
-|---------------------|---------|------------|---------------------|
-| Development         | 1 CPU   | 1 GB       | Up to 1 container   |
-| Production          | 2 CPU   | 2 GB       | Up to 10 containers |
+| **Deployment Type** | **CPU/Memory**  | **Scaling**         | **Database**                                                                     |
+|---------------------|-----------------|---------------------|----------------------------------------------------------------------------------|
+| Development         | 1 CPU, 1 GB RAM | Up to 1 container   | 10 GB disk, no backups                                                           |
+| Production          | 2 CPU, 2 GB RAM | Up to 10 containers | Autoscaling disk, automatic backups, highly available (multi-zone configuration) |
 
 CPU and memory resources are per container.
 
-!!! info "For [Cloud SaaS](../concepts/langgraph_cloud.md)"
+!!! warning "Immutable Deployment Type"
+
+    Once a deployment is created, the deployment type cannot be changed.
+
+!!! info "Resource Customization"
     For `Production` type deployments, resources can be manually increased on a case-by-case basis depending on use case and capacity constraints. Contact support@langchain.dev to request an increase in resources.
 
-!!! info 
+    For `Development` types deployments, database disk size can be manually increased on a case-by-case basis depending on use case and capacity constraints. For most use cases, [TTLs](../how-tos/ttl/configure_ttl.md) should be configured to manage disk usage. Contact support@langchain.dev to request an increase in resources.
+
     Resources for [Self-Hosted Data Plane](../concepts/langgraph_self_hosted_data_plane.md) and [Self-Hosted Control Plane](../concepts/langgraph_self_hosted_control_plane.md) deployments can be fully customized.
 
 ### Database Provisioning

docs/docs/concepts/langgraph_self_hosted_control_plane.md

@@ -2,8 +2,8 @@
 
 There are two versions of the self-hosted deployment: [Self-Hosted Data Plane](./deployment_options.md#self-hosted-data-plane) and [Self-Hosted Control Plane](./deployment_options.md#self-hosted-control-plane).
 
-!!! important "Beta"
-    The Self-Hosted Control Plane deployment option is currently in beta stage.
+!!! info "Important"
+    The Self-Hosted Control Plane deployment option is currently in beta stage and requires an [Enterprise](../../concepts/plans.md) plan.
 
 ## Requirements
 

docs/docs/concepts/langgraph_self_hosted_data_plane.md

@@ -7,8 +7,8 @@ search:
 
 There are two versions of the self-hosted deployment: [Self-Hosted Data Plane](./deployment_options.md#self-hosted-data-plane) and [Self-Hosted Control Plane](./deployment_options.md#self-hosted-control-plane).
 
-!!! important "Beta"
-    The Self-Hosted Data Plane deployment option is currently in beta stage.
+!!! info "Important"
+    The Self-Hosted Data Plane deployment option is currently in beta stage and requires an [Enterprise](../../concepts/plans.md) plan.
 
 ## Requirements
 

docs/docs/how-tos/graph-api.ipynb

@@ -2235,7 +2235,7 @@
     "    if termination_condition(state):\n",
     "        return END\n",
     "    else:\n",
-    "        return \"a\"\n",
+    "        return \"b\"\n",
     "\n",
     "builder.add_edge(START, \"a\")\n",
     "builder.add_conditional_edges(\"a\", route)\n",
@@ -2950,16 +2950,6 @@
     "    When you send updates from a subgraph node to a parent graph node for a key that's shared by both parent and subgraph [state schemas](../../concepts/low_level#schema), you **must** define a [reducer](../../concepts/low_level#reducers) for the key you're updating in the parent graph state. See the example below."
    ]
   },
-  {
-   "cell_type": "markdown",
-   "id": "6be0aeb9-e138-4adc-a1df-5d743a8eb348",
-   "metadata": {},
-   "source": [
-    "!!! important \"State updates with `Command.PARENT`\"\n",
-    "\n",
-    "    When you send updates from a subgraph node to a parent graph node for a key that's shared by both parent and subgraph [state schemas](../../concepts/low_level#schema), you **must** define a [reducer](../../concepts/low_level#reducers) for the key you're updating in the parent graph state."
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": 6,

docs/docs/stylesheets/agent_graph_widget.css

@@ -0,0 +1,57 @@
+.agent-layout {
+    display: flex;
+    flex-wrap: nowrap;
+    gap: 1rem;
+    align-items: flex-start;
+    margin-top: 1rem;
+}
+
+.agent-layout h3 {
+    margin: 0;
+}
+
+.agent-graph-features {
+    display: flex;
+    flex-direction: column;
+    gap: 0.5rem;
+    padding: 1rem;
+    max-width: 300px;
+    flex-shrink: 0;
+
+    border: 1px solid var(--md-default-fg-color--lightest);
+    border-radius: 0.5rem;
+    background-color: var(--md-default-bg-color);
+}
+
+.agent-graph-features label {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    font-size: 0.9rem;
+    color: var(--md-typeset-color);
+    cursor: pointer;
+}
+
+.agent-graph-features input[type="checkbox"] {
+    accent-color: var(--md-accent-fg-color);
+    transform: scale(1.2);
+}
+
+.agent-graph-container {
+    flex: 1 1 50%;
+    max-width: 70%;
+    padding: 1rem;
+
+    overflow: auto;
+    height: auto;
+    box-sizing: border-box;
+
+    border: 1px solid var(--md-default-fg-color--lightest);
+    border-radius: 0.5rem;
+    background-color: var(--md-default-bg-color);
+}
+
+.agent-graph-container img {
+    display: block;
+    margin: 0 auto;
+}

docs/docs/tutorials/get-started/1-build-basic-chatbot.md

@@ -6,7 +6,7 @@ In this tutorial, you will build a basic chatbot. This chatbot is the basis for
 
 Before you start this tutorial, ensure you have access to a LLM that supports
 tool-calling features, such as [OpenAI](https://platform.openai.com/api-keys),
-[Anthropic](https://console.anthropic.com/settings/admin-keys), or
+[Anthropic](https://console.anthropic.com/settings/keys), or
 [Google Gemini](https://ai.google.dev/gemini-api/docs/api-key).
 
 ## 1. Install packages

docs/docs/tutorials/get-started/2-add-tools.md

@@ -146,7 +146,7 @@ graph_builder.add_node("tools", tool_node)
 
 !!! note
 
-    If you do not want to build this yourself in the future, you can use LangGraph's prebuilt [ToolNode](https://langchain-ai.github.io/langgraph/reference/prebuilt/#toolnode).
+    If you do not want to build this yourself in the future, you can use LangGraph's prebuilt [ToolNode](https://langchain-ai.github.io/langgraph/reference/agents/#langgraph.prebuilt.tool_node.ToolNode).
 
 ## 6. Define the `conditional_edges`
 

docs/docs/tutorials/overview.md

@@ -0,0 +1,21 @@
+# Examples
+
+The pages in this section provide end-to-end examples for the following topics:
+
+## General 
+
+- [Agentic RAG](./rag/langgraph_adaptive_rag.ipynb)
+- [Agent Supervisor](./multi_agent/agent_supervisor.ipynb)
+- [SQL agent](./sql-agent.ipynb)
+- [Graph runs in LangSmith](../how-tos/run-id-langsmith.ipynb)
+
+## LangGraph Platform
+
+- [Set up custom authentication](./auth/getting_started.md)
+- [Make conversations private](./auth/resource_auth.md)
+- [Connect an authentication provider](./auth/add_auth_server.md)
+- [Rebuild graph at runtime](../cloud/deployment/graph_rebuild.md)
+- [Use RemoteGraph](../how-tos/use-remote-graph.md)
+- [Deploy CrewAI, AutoGen, and other frameworks](../how-tos/autogen-langgraph-platform.ipynb)
+- [Integrate LangGraph into a React app](../cloud/how-tos/use_stream_react.md)
+- [Implement Generative User Interfaces with LangGraph](../cloud/how-tos/generative_ui_react.md)

docs/mkdocs.yml

@@ -19,7 +19,6 @@ theme:
     - content.tabs.link
     - content.action.edit
     - content.tooltips
-    - header.autohide
     - navigation.indexes
     - navigation.footer
     - navigation.instant
@@ -28,7 +27,6 @@ theme:
     - navigation.instant.progress
     - navigation.path
     - navigation.tabs
-    - navigation.tabs.sticky
     - navigation.top
     - navigation.prune
     - navigation.tracking
@@ -418,3 +416,5 @@ extra_css:
   - stylesheets/version_admonitions.css
   - stylesheets/logos.css
   - stylesheets/sticky_navigation.css
+  - stylesheets/agent_graph_widget.css
+

docs/uv.lock

@@ -2590,7 +2590,7 @@ wheels = [
 
 [[package]]
 name = "langgraph"
-version = "0.4.5"
+version = "0.4.7"
 source = { editable = "../libs/langgraph" }
 dependencies = [
     { name = "langchain-core" },
@@ -2891,7 +2891,7 @@ test = [
 
 [[package]]
 name = "langgraph-prebuilt"
-version = "0.1.8"
+version = "0.2.2"
 source = { editable = "../libs/prebuilt" }
 dependencies = [
     { name = "langchain-core" },
@@ -2917,11 +2917,12 @@ dev = [
     { name = "pytest-mock" },
     { name = "pytest-watcher" },
     { name = "ruff" },
+    { name = "syrupy" },
 ]
 
 [[package]]
 name = "langgraph-sdk"
-version = "0.1.69"
+version = "0.1.70"
 source = { editable = "../libs/sdk-py" }
 dependencies = [
     { name = "httpx" },
@@ -2947,8 +2948,8 @@ dev = [
 
 [[package]]
 name = "langgraph-supervisor"
-version = "0.0.21"
-source = { git = "https://github.com/langchain-ai/langgraph-supervisor-py#6367bebd5462ac899e7def931ac6ab9cc6a9b070" }
+version = "0.0.25"
+source = { git = "https://github.com/langchain-ai/langgraph-supervisor-py#79380b5c21d3170e2d20dc6c55149ee057a306b1" }
 dependencies = [
     { name = "langchain-core" },
     { name = "langgraph" },

libs/langgraph/README.md

@@ -74,7 +74,7 @@ While LangGraph can be used standalone, it also integrates seamlessly with any L
 
 - [Guides](https://langchain-ai.github.io/langgraph/how-tos/): Quick, actionable code snippets for topics such as streaming, adding memory & persistence, and design patterns (e.g. branching, subgraphs, etc.).
 - [Reference](https://langchain-ai.github.io/langgraph/reference/graphs/): Detailed reference on core classes, methods, how to use the graph and checkpointing APIs, and higher-level prebuilt components.
-- [Examples](https://langchain-ai.github.io/langgraph/tutorials/): Guided examples on getting started with LangGraph.
+- [Examples](https://langchain-ai.github.io/langgraph/tutorials/overview/): Guided examples on getting started with LangGraph.
 - [LangChain Academy](https://academy.langchain.com/courses/intro-to-langgraph): Learn the basics of LangGraph in our free, structured course.
 - [Templates](https://langchain-ai.github.io/langgraph/concepts/template_applications/): Pre-built reference apps for common agentic workflows (e.g. ReAct agent, memory, retrieval etc.) that can be cloned and adapted.
 - [Case studies](https://www.langchain.com/built-with-langgraph): Hear how industry leaders use LangGraph to ship AI applications at scale.

libs/langgraph/langgraph/graph/message.py

@@ -23,6 +23,7 @@ from langchain_core.messages import (
 )
 from typing_extensions import TypedDict
 
+from langgraph.constants import CONF, CONFIG_KEY_SEND
 from langgraph.graph.state import StateGraph
 
 Messages = Union[list[MessageLikeRepresentation], MessageLikeRepresentation]
@@ -294,3 +295,52 @@ def _format_messages(messages: Sequence[BaseMessage]) -> list[BaseMessage]:
         return list(messages)
     else:
         return convert_to_messages(convert_to_openai_messages(messages))
+
+
+def push_message(
+    message: Union[MessageLikeRepresentation, BaseMessageChunk],
+    *,
+    state_key: Optional[str] = "messages",
+) -> AnyMessage:
+    """Write a message manually to the `messages` / `messages-tuple` stream mode.
+
+    Will automatically write to the channel specified in the `state_key` unless `state_key` is `None`.
+    """
+
+    from langchain_core.callbacks.base import (
+        BaseCallbackHandler,
+        BaseCallbackManager,
+    )
+
+    from langgraph.config import get_config
+    from langgraph.constants import NS_SEP
+    from langgraph.pregel.messages import StreamMessagesHandler
+
+    config = get_config()
+    message = next(x for x in convert_to_messages([message]))
+
+    if message.id is None:
+        raise ValueError("Message ID is required")
+
+    if isinstance(config["callbacks"], BaseCallbackManager):
+        manager = config["callbacks"]
+        handlers = manager.handlers
+    elif isinstance(config["callbacks"], list) and all(
+        isinstance(x, BaseCallbackHandler) for x in config["callbacks"]
+    ):
+        handlers = config["callbacks"]
+
+    if stream_handler := next(
+        (x for x in handlers if isinstance(x, StreamMessagesHandler)), None
+    ):
+        metadata = config["metadata"]
+        message_meta = (
+            tuple(cast(str, metadata["langgraph_checkpoint_ns"]).split(NS_SEP)),
+            metadata,
+        )
+        stream_handler._emit(message_meta, message, dedupe=False)
+
+    if state_key:
+        config[CONF][CONFIG_KEY_SEND]([(state_key, message)])
+
+    return message

libs/langgraph/langgraph/graph/schema_utils.py

@@ -1,268 +0,0 @@
-import functools
-import logging
-import weakref
-from dataclasses import is_dataclass
-from inspect import isclass
-from typing import (
-    Annotated,
-    Any,
-    Callable,
-    Optional,
-    Union,
-    get_args,
-    get_origin,
-    get_type_hints,
-)
-
-from pydantic import BaseModel, ConfigDict, TypeAdapter
-from typing_extensions import is_typeddict
-
-__all__ = ["SchemaCoercionMapper"]
-
-logger = logging.getLogger(__name__)
-
-
-_cache: weakref.WeakKeyDictionary[type[Any], dict[int, "SchemaCoercionMapper"]] = (
-    weakref.WeakKeyDictionary()
-)
-
-
-class SchemaCoercionMapper:
-    """Lightweight coercion of *dict* → *BaseModel* instances."""
-
-    def __new__(
-        cls,
-        schema: type[Any],
-        type_hints: Optional[dict[str, Any]] = None,
-        *,
-        max_depth: int = 12,
-    ) -> "SchemaCoercionMapper":
-        by_depth = _cache.setdefault(schema, {})
-        if max_depth in by_depth:
-            return by_depth[max_depth]
-        inst = super().__new__(cls)
-        by_depth[max_depth] = inst
-        return inst
-
-    def __init__(
-        self,
-        schema: type[BaseModel],
-        type_hints: Optional[dict[str, Any]] = None,
-        *,
-        max_depth: int = 12,
-    ) -> None:
-        if hasattr(self, "_initialised"):
-            return
-        self._initialised = True
-
-        self.schema = schema
-        self.max_depth = max_depth
-
-        self.type_hints = (
-            type_hints
-            if type_hints is not None
-            else get_type_hints(schema, localns={schema.__name__: schema})
-        )
-
-        if issubclass(schema, BaseModel):
-            self._fields = {
-                n: self.type_hints.get(n, f.annotation)
-                for n, f in schema.model_fields.items()
-            }
-            self._construct: Callable[..., Any] = schema.model_construct
-            unhandled_attrs = ("validators", "field_validators", "root_validators")
-            if (decorators := getattr(schema, "__pydantic_decorators__", None)) and any(
-                getattr(decorators, attr, None) for attr in unhandled_attrs
-            ):
-                self.coerce = lambda v, _: schema.model_validate(v)
-            else:
-                self.coerce = self._coerce
-        else:
-            raise TypeError("Schema must be a Pydantic V2 model.")
-
-        self._field_coercers: Optional[dict[str, Callable[[Any, int], Any]]] = None
-
-    def __call__(self, input_data: Any, depth: Optional[int] = None) -> Any:
-        return self.coerce(input_data, depth)
-
-    def _coerce(self, input_data: Any, depth: Optional[int] = None) -> Any:
-        if depth is None:
-            depth = self.max_depth
-        if not isinstance(input_data, dict) or depth <= 0:
-            return input_data
-
-        if self._field_coercers is None:
-            self._field_coercers = {
-                n: self._build_coercer(t, depth - 1) for n, t in self._fields.items()
-            }
-
-        processed: dict[str, Any] = {}
-        for k, v in input_data.items():
-            fn = self._field_coercers.get(k)
-            processed[k] = fn(v, depth - 1) if fn else v
-        return self._construct(**processed)
-
-    def _build_coercer(
-        self, field_type: Any, depth: int, *, throw: bool = False
-    ) -> Callable[[Any, Any], Any]:
-        if depth == 0:
-            return self._passthrough
-
-        origin = get_origin(field_type)
-
-        if (field_type in _IDENTITY_TYPES) or (origin in _IDENTITY_TYPES):
-            return self._passthrough
-
-        if origin is Annotated:
-            real_type, *_ = get_args(field_type)
-            sub = self._build_coercer(real_type, depth - 1)
-            return lambda v, d: sub(v, d)
-
-        if isclass(field_type):
-            # This is needed bcs. of issubclass issues on older versions of python
-            try:
-                is_bm_subclass = issubclass(field_type, BaseModel)
-            except TypeError:
-                # python < 3.11 issue.
-                is_bm_subclass = False
-            if is_bm_subclass:
-                mapper = SchemaCoercionMapper(field_type, max_depth=depth - 1)
-                return lambda v, d: mapper.coerce(v, d) if isinstance(v, dict) else v
-
-        if origin is list:
-            args = get_args(field_type)
-            if len(args) != 1:
-                return self._passthrough
-            sub = self._build_coercer(args[0], depth - 1)
-
-            def list_coercer(v: Any, d: Any) -> Any:
-                if not isinstance(v, (list, tuple)):
-                    return v
-                return [sub(x, d - 1) for x in v]
-
-            return list_coercer
-
-        if origin is set or field_type is set:
-            args = get_args(field_type)
-            if len(args) > 1:
-                return self._passthrough
-            elif len(args) == 1:
-                sub = self._build_coercer(args[0], depth - 1)
-            else:
-                sub = None  # type: ignore
-
-            def set_coercer(v: Any, d: Any) -> Any:
-                if not isinstance(v, (list, tuple, set)):
-                    return v
-                if sub is None:
-                    return set(v)
-                return {sub(x, d - 1) for x in v}
-
-            return set_coercer
-        if origin is dict or field_type is dict:
-            args = get_args(field_type)
-            if len(args) != 2:
-
-                def dict_coercer(v: Any, d: Any) -> Any:
-                    if not isinstance(v, dict):
-                        if throw:
-                            raise TypeError(f"Expected dict, got {type(v)}")
-                    return v
-
-                return dict_coercer
-            k_sub = self._build_coercer(args[0], depth - 1)
-            v_sub = self._build_coercer(args[1], depth - 1)
-
-            def dict_coercer(v: Any, d: Any) -> Any:
-                if not isinstance(v, dict):
-                    if throw:
-                        raise TypeError(f"Expected dict, got {type(v)}")
-                    return v
-                return {k_sub(k, d - 1): v_sub(val, d - 1) for k, val in v.items()}
-
-            return dict_coercer
-
-        if origin is tuple:
-            elem_types = get_args(field_type)
-            if not elem_types:
-                return self._passthrough
-            subs = [self._build_coercer(t, depth - 1) for t in elem_types]
-            return lambda v, d: (
-                tuple(
-                    subs[i](v[i] if i < len(v) else None, d - 1)
-                    for i in range(len(subs))
-                )
-                if isinstance(v, (list, tuple))
-                else v
-            )
-
-        if origin is Union:
-            uargs = get_args(field_type)
-            subs, none_in_union = [], False
-            for ix, arg in enumerate(uargs):
-                if arg is type(None):
-                    none_in_union = True
-                else:
-                    subs.append(
-                        self._build_coercer(arg, depth - 1, throw=ix < len(uargs) - 1)
-                    )
-
-            def union_coercer(v: Any, d: Any) -> Any:
-                if v is None and none_in_union:
-                    return None
-                err = None
-                for sp in subs:
-                    try:
-                        return sp(v, d - 1)
-                    except TypeError as e:
-                        err = e
-                if err:
-                    raise err
-                return v
-
-            return union_coercer
-
-        adapter_fn = _get_adapter(field_type)
-        return lambda v, _d: adapter_fn(v)
-
-    @staticmethod
-    def _passthrough(v: Any, _d: Any) -> Any:  # noqa: D401
-        return v
-
-
-_adapter_cache: dict[Any, Callable[[Any], Any]] = {}
-
-
-_IDENTITY_TYPES: tuple[type[Any], ...] = (
-    int,
-    float,
-    str,
-    bool,
-    bytes,
-    bytearray,
-    complex,
-    memoryview,
-    type(None),
-)
-
-
-@functools.lru_cache(maxsize=2048)
-def _adapter_for(tp: Any) -> Callable[[Any], Any]:  # noqa: D401
-    try:
-        config = (
-            None
-            if (issubclass(tp, BaseModel) or is_dataclass(tp) or is_typeddict(tp))
-            else ConfigDict(arbitrary_types_allowed=True)
-        )
-    except TypeError:
-        config = None
-    return TypeAdapter(tp, config=config).validate_python
-
-
-def _get_adapter(tp: Any) -> Callable[[Any], Any]:
-    try:
-        return _adapter_cache[tp]
-    except KeyError:
-        fn = _adapter_for(tp)
-        _adapter_cache[tp] = fn
-        return fn

libs/langgraph/langgraph/graph/state.py

@@ -64,7 +64,6 @@ from langgraph.graph.graph import (
     Graph,
     Send,
 )
-from langgraph.graph.schema_utils import SchemaCoercionMapper
 from langgraph.managed.base import (
     ChannelKeyPlaceholder,
     ChannelTypePlaceholder,
@@ -1042,11 +1041,8 @@ def _pick_mapper(
 ) -> Optional[Callable[[Any], Any]]:
     if state_keys == ["__root__"]:
         return None
-    if isclass(schema):
-        if issubclass(schema, dict):
-            return None
-        if issubclass(schema, BaseModel):
-            return SchemaCoercionMapper(schema, type_hints=type_hints)
+    if isclass(schema) and issubclass(schema, dict):
+        return None
     return partial(_coerce_state, schema)
 
 

libs/langgraph/langgraph/graph/ui.py

@@ -54,7 +54,7 @@ def push_ui_message(
     id: Optional[str] = None,
     metadata: Optional[dict[str, Any]] = None,
     message: Optional[AnyMessage] = None,
-    state_key: str = "ui",
+    state_key: Optional[str] = "ui",
     merge: bool = False,
 ) -> UIMessage:
     """Push a new UI message to update the UI state.
@@ -111,7 +111,8 @@ def push_ui_message(
     }
 
     writer(evt)
-    config[CONF][CONFIG_KEY_SEND]([(state_key, evt)])
+    if state_key:
+        config[CONF][CONFIG_KEY_SEND]([(state_key, evt)])
 
     return evt
 

libs/langgraph/langgraph/pregel/__init__.py

@@ -2214,12 +2214,14 @@ class Pregel(PregelProtocol):
             validate_keys(output_keys, self.channels)
         interrupt_before = interrupt_before or self.interrupt_before_nodes
         interrupt_after = interrupt_after or self.interrupt_after_nodes
-        stream_mode = stream_mode if stream_mode is not None else self.stream_mode
+        if stream_mode is None and CONFIG_KEY_TASK_ID in config.get(CONF, {}):
+            # if being called as a node in another graph, default to values mode
+            # but don't overwrite stream_mode arg if provided
+            stream_mode = ["values"]
+        elif stream_mode is None:
+            stream_mode = self.stream_mode
         if not isinstance(stream_mode, list):
             stream_mode = [stream_mode]
-        if CONFIG_KEY_TASK_ID in config.get(CONF, {}):
-            # if being called as a node in another graph, always use values mode
-            stream_mode = ["values"]
         if self.checkpointer is False:
             checkpointer: BaseCheckpointSaver | None = None
         elif CONFIG_KEY_CHECKPOINTER in config.get(CONF, {}):

libs/langgraph/langgraph/pregel/loop.py

@@ -135,7 +135,6 @@ P = ParamSpec("P")
 INPUT_DONE = object()
 INPUT_RESUMING = object()
 INPUT_SHOULD_VALIDATE = object()
-SPECIAL_CHANNELS = (ERROR, INTERRUPT, SCHEDULED)
 WritesT = Sequence[tuple[str, Any]]
 
 
@@ -892,7 +891,7 @@ class PregelLoop(LoopProtocol):
                 and self.checkpoint_pending_writes
                 and any(task.writes for task in self.tasks.values())
             ):
-                mv_writes, _ = apply_writes(
+                mv_writes, updated_channels = apply_writes(
                     self.checkpoint,
                     self.channels,
                     self.tasks.values(),
@@ -901,13 +900,18 @@ class PregelLoop(LoopProtocol):
                 )
                 for key, values in mv_writes.items():
                     self._update_mv(key, values)
-                self._emit(
-                    "values",
-                    map_output_values,
-                    self.output_keys,
-                    [w for t in self.tasks.values() for w in t.writes],
-                    self.channels,
-                )
+                if not updated_channels.isdisjoint(
+                    (self.output_keys,)
+                    if isinstance(self.output_keys, str)
+                    else self.output_keys
+                ):
+                    self._emit(
+                        "values",
+                        map_output_values,
+                        self.output_keys,
+                        [w for t in self.tasks.values() for w in t.writes],
+                        self.channels,
+                    )
             # emit INTERRUPT if exception is empty (otherwise emitted by put_writes)
             if exc_value is not None and (not exc_value.args or not exc_value.args[0]):
                 self._emit(

libs/langgraph/langgraph/pregel/retry.py

@@ -41,7 +41,7 @@ def run_with_retry(
         except ParentCommand as exc:
             ns: str = config[CONF][CONFIG_KEY_CHECKPOINT_NS]
             cmd = exc.args[0]
-            if cmd.graph == ns:
+            if cmd.graph in (ns, task.name):
                 # this command is for the current graph, handle it
                 for w in task.writers:
                     w.invoke(cmd, config)
@@ -137,7 +137,7 @@ async def arun_with_retry(
         except ParentCommand as exc:
             ns: str = config[CONF][CONFIG_KEY_CHECKPOINT_NS]
             cmd = exc.args[0]
-            if cmd.graph == ns:
+            if cmd.graph in (ns, task.name):
                 # this command is for the current graph, handle it
                 for w in task.writers:
                     w.invoke(cmd, config)

libs/langgraph/langgraph/pregel/runner.py

@@ -56,6 +56,10 @@ EXCLUDED_FRAME_FNAMES = (
     "concurrent/futures/_base.py",
 )
 
+SKIP_RERAISE_SET: weakref.WeakSet[Union[concurrent.futures.Future, asyncio.Future]] = (
+    weakref.WeakSet()
+)
+
 
 class FuturesDict(Generic[F, E], dict[F, Optional[PregelExecutableTask]]):
     event: E
@@ -165,7 +169,6 @@ class PregelRunner:
                             futures=weakref.ref(futures),
                             schedule_task=schedule_task,
                             submit=self.submit,
-                            reraise=reraise,
                         ),
                     },
                 )
@@ -207,7 +210,6 @@ class PregelRunner:
                         futures=weakref.ref(futures),
                         schedule_task=schedule_task,
                         submit=self.submit,
-                        reraise=reraise,
                     ),
                 },
                 __reraise_on_exit__=reraise,
@@ -302,7 +304,6 @@ class PregelRunner:
                             futures=weakref.ref(futures),
                             schedule_task=schedule_task,
                             submit=self.submit,
-                            reraise=reraise,
                             loop=loop,
                         ),
                     },
@@ -349,7 +350,6 @@ class PregelRunner:
                             futures=weakref.ref(futures),
                             schedule_task=schedule_task,
                             submit=self.submit,
-                            reraise=reraise,
                             loop=loop,
                         ),
                     },
@@ -434,7 +434,8 @@ class PregelRunner:
                 raise exception
             else:
                 # save error to checkpointer
-                self.put_writes()(task.id, [(ERROR, exception)])  # type: ignore[misc]
+                task.writes.append((ERROR, exception))
+                self.put_writes()(task.id, task.writes)  # type: ignore[misc]
         else:
             if self.node_finished and (
                 task.config is None or TAG_HIDDEN not in task.config.get("tags", [])
@@ -456,7 +457,7 @@ def _should_stop_others(
         if fut.cancelled():
             continue
         elif exc := fut.exception():
-            if not isinstance(exc, GraphBubbleUp):
+            if not isinstance(exc, GraphBubbleUp) and fut not in SKIP_RERAISE_SET:
                 return True
 
     return False
@@ -494,7 +495,8 @@ def _panic_or_proceed(
     interrupts: list[GraphInterrupt] = []
     while done:
         # if any task failed
-        if exc := _exception(done.pop()):
+        fut = done.pop()
+        if exc := _exception(fut):
             # cancel all pending tasks
             while inflight:
                 inflight.pop().cancel()
@@ -503,7 +505,7 @@ def _panic_or_proceed(
                 if isinstance(exc, GraphInterrupt):
                     # collect interrupts
                     interrupts.append(exc)
-                else:
+                elif fut not in SKIP_RERAISE_SET:
                     raise exc
     # raise combined interrupts
     if interrupts:
@@ -530,7 +532,6 @@ def _call(
         [PregelExecutableTask, int, Optional[Call]], Optional[PregelExecutableTask]
     ],
     submit: weakref.ref[Submit],
-    reraise: bool,
 ) -> concurrent.futures.Future[Any]:
     if asyncio.iscoroutinefunction(func):
         raise RuntimeError("In an sync context async tasks cannot be called")
@@ -582,14 +583,16 @@ def _call(
                         callbacks=callbacks,
                         schedule_task=schedule_task,
                         submit=submit,
-                        reraise=reraise,
                     ),
                 },
-                __reraise_on_exit__=reraise,
+                __reraise_on_exit__=False,
                 # starting a new task in the next tick ensures
                 # updates from this tick are committed/streamed first
                 __next_tick__=True,
             )
+            # exceptions for call() tasks are raised into the parent task
+            # so we should not re-raise at the end of the tick
+            SKIP_RERAISE_SET.add(fut)
             futures()[fut] = next_task  # type: ignore[index]
     fut = cast(Union[asyncio.Future, concurrent.futures.Future], fut)
     # return a chained future to ensure commit() callback is called
@@ -613,7 +616,6 @@ def _acall(
     ],
     submit: weakref.ref[Submit],
     loop: asyncio.AbstractEventLoop,
-    reraise: bool = False,
     stream: bool = False,
 ) -> Union[asyncio.Future[Any], concurrent.futures.Future[Any]]:
     # return a chained future to ensure commit() callback is called
@@ -643,7 +645,6 @@ def _acall(
             schedule_task=schedule_task,
             submit=submit,
             loop=loop,
-            reraise=reraise,
             stream=stream,
         ),
         loop,
@@ -669,7 +670,6 @@ async def _acall_impl(
     ],
     submit: weakref.ref[Submit],
     loop: asyncio.AbstractEventLoop,
-    reraise: bool = False,
     stream: bool = False,
 ) -> None:
     try:
@@ -726,17 +726,19 @@ async def _acall_impl(
                                 schedule_task=schedule_task,
                                 submit=submit,
                                 loop=loop,
-                                reraise=reraise,
                             ),
                         },
-                        __name__=task().name,  # type: ignore[union-attr]
+                        __name__=next_task.name,
                         __cancel_on_exit__=True,
-                        __reraise_on_exit__=reraise,
+                        __reraise_on_exit__=False,
                         # starting a new task in the next tick ensures
                         # updates from this tick are committed/streamed first
                         __next_tick__=True,
                     ),
                 )
+                # exceptions for call() tasks are raised into the parent task
+                # so we should not re-raise at the end of the tick
+                SKIP_RERAISE_SET.add(fut)
                 futures()[fut] = next_task  # type: ignore[index]
         if fut is not None:
             chain_future(fut, destination)

libs/langgraph/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "langgraph"
-version = "0.4.5"
+version = "0.4.8"
 description = "Building stateful, multi-actor applications with LLMs"
 authors = []
 requires-python = ">=3.9"
@@ -15,7 +15,7 @@ dependencies = [
     "langchain-core>=0.1",
     "langgraph-checkpoint>=2.0.26",
     "langgraph-sdk>=0.1.42",
-    "langgraph-prebuilt>=0.1.8",
+    "langgraph-prebuilt>=0.2.0",
     "xxhash>=3.5.0",
     "pydantic>=2.7.4",
 ]

libs/langgraph/tests/test_messages_state.py

@@ -15,7 +15,7 @@ from pydantic import BaseModel
 from typing_extensions import TypedDict
 
 from langgraph.graph import add_messages
-from langgraph.graph.message import REMOVE_ALL_MESSAGES, MessagesState
+from langgraph.graph.message import REMOVE_ALL_MESSAGES, MessagesState, push_message
 from langgraph.graph.state import END, START, StateGraph
 from tests.messages import _AnyIdHumanMessage
 
@@ -332,3 +332,34 @@ def test_remove_all_messages():
     assert result == [
         _AnyIdHumanMessage(content="Updated hi there"),
     ]
+
+
+def test_push_messages_in_graph():
+    class MessagesState(TypedDict):
+        messages: Annotated[list[AnyMessage], add_messages]
+
+    def chat(_: MessagesState) -> MessagesState:
+        with pytest.raises(ValueError, match="Message ID is required"):
+            push_message(AIMessage(content="No ID"))
+
+        push_message(AIMessage(content="First", id="1"))
+        push_message(HumanMessage(content="Second", id="2"))
+        push_message(AIMessage(content="Third", id="3"))
+
+    builder = StateGraph(MessagesState)
+    builder.add_node(chat)
+    builder.add_edge(START, "chat")
+
+    graph = builder.compile()
+
+    messages, values = [], None
+    for event, chunk in graph.stream(
+        {"messages": []}, stream_mode=["messages", "values"]
+    ):
+        if event == "values":
+            values = chunk
+        elif event == "messages":
+            message, _ = chunk
+            messages.append(message)
+
+    assert values["messages"] == messages

libs/langgraph/tests/test_pregel.py

@@ -5514,8 +5514,11 @@ def test_runnable_passthrough_node_graph() -> None:
     assert graph.get_graph(xray=True).to_json() == graph.get_graph(xray=False).to_json()
 
 
+@pytest.mark.parametrize("subgraph_persist", [True, False])
 @pytest.mark.parametrize("checkpointer_name", ALL_CHECKPOINTERS_SYNC)
-def test_parent_command(request: pytest.FixtureRequest, checkpointer_name: str) -> None:
+def test_parent_command(
+    request: pytest.FixtureRequest, checkpointer_name: str, subgraph_persist: bool
+) -> None:
     from langchain_core.messages import BaseMessage
     from langchain_core.tools import tool
 
@@ -5527,7 +5530,7 @@ def test_parent_command(request: pytest.FixtureRequest, checkpointer_name: str)
     subgraph_builder = StateGraph(MessagesState)
     subgraph_builder.add_node("tool", get_user_name)
     subgraph_builder.add_edge(START, "tool")
-    subgraph = subgraph_builder.compile()
+    subgraph = subgraph_builder.compile(checkpointer=subgraph_persist)
 
     class CustomParentState(TypedDict):
         messages: Annotated[list[BaseMessage], add_messages]
@@ -6873,7 +6876,7 @@ def test_sync_streaming_with_functional_api() -> None:
     should be greater than the time delay between the two tasks.
     """
 
-    time_delay = 0.01
+    time_delay = 0.05
 
     @task()
     def slow() -> dict:
@@ -8769,3 +8772,76 @@ def test_get_graph_root_channel(snapshot: SnapshotAssertion) -> None:
 
     assert json.dumps(graph.get_graph().to_json(), indent=2) == snapshot
     assert graph.get_graph().draw_mermaid(with_styles=False) == snapshot
+
+
+def test_imp_exception(
+    sync_checkpointer: BaseCheckpointSaver,
+) -> None:
+    @task()
+    def my_task(number: int):
+        time.sleep(0.1)
+        return number * 2
+
+    @task()
+    def task_with_exception(number: int):
+        time.sleep(0.1)
+        raise Exception("This is a test exception")
+
+    @entrypoint(checkpointer=sync_checkpointer)
+    def my_workflow(number: int):
+        my_task(number).result()
+        try:
+            task_with_exception(number).result()
+        except Exception as e:
+            print(f"Exception caught: {e}")
+        my_task(number).result()
+        return "done"
+
+    thread1 = {"configurable": {"thread_id": "1"}}
+    assert my_workflow.invoke(1, thread1) == "done"
+
+    assert [c for c in my_workflow.stream(1, thread1)] == [
+        {"my_task": 2},
+        {"my_task": 2},
+        {"my_workflow": "done"},
+    ]
+
+
+@pytest.mark.parametrize("subgraph_persist", [True, False])
+def test_parent_command_goto(
+    sync_checkpointer: BaseCheckpointSaver, subgraph_persist: bool
+) -> None:
+    class State(TypedDict):
+        dialog_state: Annotated[list[str], operator.add]
+
+    def node_a_child(state):
+        return {"dialog_state": ["a_child_state"]}
+
+    def node_b_child(state):
+        return Command(
+            graph=Command.PARENT,
+            goto="node_b_parent",
+            update={"dialog_state": ["b_child_state"]},
+        )
+
+    sub_builder = StateGraph(State)
+    sub_builder.add_node(node_a_child)
+    sub_builder.add_node(node_b_child)
+    sub_builder.add_edge(START, "node_a_child")
+    sub_builder.add_edge("node_a_child", "node_b_child")
+    sub_graph = sub_builder.compile(checkpointer=subgraph_persist)
+
+    def node_b_parent(state):
+        return {"dialog_state": ["node_b_parent"]}
+
+    main_builder = StateGraph(State)
+    main_builder.add_node(node_b_parent)
+    main_builder.add_edge(START, "subgraph_node")
+    main_builder.add_node("subgraph_node", sub_graph, destinations=("node_b_parent",))
+
+    main_graph = main_builder.compile(sync_checkpointer, name="parent")
+    config = {"configurable": {"thread_id": 1}}
+
+    assert main_graph.invoke(input={"dialog_state": ["init_state"]}, config=config) == {
+        "dialog_state": ["init_state", "b_child_state", "node_b_parent"]
+    }

libs/langgraph/tests/test_pregel_async.py

@@ -6772,8 +6772,9 @@ async def test_debug_nested_subgraphs(async_checkpointer: BaseCheckpointSaver):
                 assert stream_task.get("state") == history_task.state
 
 
+@pytest.mark.parametrize("subgraph_persist", [True, False])
 @pytest.mark.parametrize("checkpointer_name", ALL_CHECKPOINTERS_ASYNC)
-async def test_parent_command(checkpointer_name: str) -> None:
+async def test_parent_command(checkpointer_name: str, subgraph_persist: bool) -> None:
     from langchain_core.messages import BaseMessage
     from langchain_core.tools import tool
 
@@ -6785,7 +6786,7 @@ async def test_parent_command(checkpointer_name: str) -> None:
     subgraph_builder = StateGraph(MessagesState)
     subgraph_builder.add_node("tool", get_user_name)
     subgraph_builder.add_edge(START, "tool")
-    subgraph = subgraph_builder.compile()
+    subgraph = subgraph_builder.compile(checkpointer=subgraph_persist)
 
     class CustomParentState(TypedDict):
         messages: Annotated[list[BaseMessage], add_messages]
@@ -9148,3 +9149,341 @@ async def test_draw_invalid():
             {"source": "nothing", "target": "__end__"},
         ],
     }
+
+
+@NEEDS_CONTEXTVARS
+async def test_imp_exception(
+    async_checkpointer: BaseCheckpointSaver,
+) -> None:
+    @task()
+    async def my_task(number: int):
+        await asyncio.sleep(0.1)
+        return number * 2
+
+    @task()
+    async def task_with_exception(number: int):
+        await asyncio.sleep(0.1)
+        raise Exception("This is a test exception")
+
+    @entrypoint(checkpointer=async_checkpointer)
+    async def my_workflow(number: int):
+        await my_task(number)
+        try:
+            await task_with_exception(number)
+        except Exception as e:
+            print(f"Exception caught: {e}")
+        await my_task(number)
+        return "done"
+
+    thread1 = {"configurable": {"thread_id": "1"}}
+    assert await my_workflow.ainvoke(1, thread1) == "done"
+
+    assert [c async for c in my_workflow.astream(1, thread1)] == [
+        {"my_task": 2},
+        {"my_task": 2},
+        {"my_workflow": "done"},
+    ]
+
+    assert [c async for c in my_workflow.astream_events(1, thread1)] == [
+        {
+            "event": "on_chain_start",
+            "data": {"input": 1},
+            "name": "LangGraph",
+            "tags": [],
+            "run_id": AnyStr(),
+            "metadata": {"thread_id": "1"},
+            "parent_ids": [],
+        },
+        {
+            "event": "on_chain_start",
+            "data": {"input": 1},
+            "name": "my_workflow",
+            "tags": ["graph:step:4"],
+            "run_id": AnyStr(),
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_workflow",
+                "langgraph_triggers": ("__start__",),
+                "langgraph_path": ("__pregel_pull", "my_workflow"),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [AnyStr()],
+        },
+        {
+            "event": "on_chain_start",
+            "data": {"input": {"number": 1}},
+            "name": "my_task",
+            "tags": ["seq:step:1"],
+            "run_id": AnyStr(),
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_stream",
+            "run_id": AnyStr(),
+            "name": "my_task",
+            "tags": ["seq:step:1"],
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "data": {"chunk": 2},
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_end",
+            "data": {"output": 2, "input": {"number": 1}},
+            "run_id": AnyStr(),
+            "name": "my_task",
+            "tags": ["seq:step:1"],
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_stream",
+            "run_id": AnyStr(),
+            "name": "LangGraph",
+            "tags": [],
+            "metadata": {"thread_id": "1"},
+            "data": {"chunk": {"my_task": 2}},
+            "parent_ids": [],
+        },
+        {
+            "event": "on_chain_start",
+            "data": {"input": {"number": 1}},
+            "name": "task_with_exception",
+            "tags": ["seq:step:1"],
+            "run_id": AnyStr(),
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_start",
+            "data": {"input": {"number": 1}},
+            "name": "my_task",
+            "tags": ["seq:step:1"],
+            "run_id": AnyStr(),
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_stream",
+            "run_id": AnyStr(),
+            "name": "my_task",
+            "tags": ["seq:step:1"],
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "data": {"chunk": 2},
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_end",
+            "data": {"output": 2, "input": {"number": 1}},
+            "run_id": AnyStr(),
+            "name": "my_task",
+            "tags": ["seq:step:1"],
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_task",
+                "langgraph_triggers": ("__pregel_push",),
+                "langgraph_path": (
+                    "__pregel_push",
+                    ("__pregel_pull", "my_workflow"),
+                    2,
+                    True,
+                ),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [
+                AnyStr(),
+                AnyStr(),
+            ],
+        },
+        {
+            "event": "on_chain_stream",
+            "run_id": AnyStr(),
+            "name": "my_workflow",
+            "tags": ["graph:step:4"],
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_workflow",
+                "langgraph_triggers": ("__start__",),
+                "langgraph_path": ("__pregel_pull", "my_workflow"),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "data": {"chunk": "done"},
+            "parent_ids": [AnyStr()],
+        },
+        {
+            "event": "on_chain_stream",
+            "run_id": AnyStr(),
+            "name": "LangGraph",
+            "tags": [],
+            "metadata": {"thread_id": "1"},
+            "data": {"chunk": {"my_task": 2}},
+            "parent_ids": [],
+        },
+        {
+            "event": "on_chain_end",
+            "data": {"output": "done", "input": 1},
+            "run_id": AnyStr(),
+            "name": "my_workflow",
+            "tags": ["graph:step:4"],
+            "metadata": {
+                "thread_id": "1",
+                "langgraph_step": 4,
+                "langgraph_node": "my_workflow",
+                "langgraph_triggers": ("__start__",),
+                "langgraph_path": ("__pregel_pull", "my_workflow"),
+                "langgraph_checkpoint_ns": AnyStr(),
+            },
+            "parent_ids": [AnyStr()],
+        },
+        {
+            "event": "on_chain_stream",
+            "run_id": AnyStr(),
+            "name": "LangGraph",
+            "tags": [],
+            "metadata": {"thread_id": "1"},
+            "data": {"chunk": {"my_workflow": "done"}},
+            "parent_ids": [],
+        },
+        {
+            "event": "on_chain_end",
+            "data": {"output": "done"},
+            "run_id": AnyStr(),
+            "name": "LangGraph",
+            "tags": [],
+            "metadata": {"thread_id": "1"},
+            "parent_ids": [],
+        },
+    ]
+
+
+@pytest.mark.parametrize("subgraph_persist", [True, False])
+async def test_parent_command_goto(
+    async_checkpointer: BaseCheckpointSaver, subgraph_persist: bool
+) -> None:
+    class State(TypedDict):
+        dialog_state: Annotated[list[str], operator.add]
+
+    async def node_a_child(state):
+        return {"dialog_state": ["a_child_state"]}
+
+    async def node_b_child(state):
+        return Command(
+            graph=Command.PARENT,
+            goto="node_b_parent",
+            update={"dialog_state": ["b_child_state"]},
+        )
+
+    sub_builder = StateGraph(State)
+    sub_builder.add_node(node_a_child)
+    sub_builder.add_node(node_b_child)
+    sub_builder.add_edge(START, "node_a_child")
+    sub_builder.add_edge("node_a_child", "node_b_child")
+    sub_graph = sub_builder.compile(checkpointer=subgraph_persist)
+
+    async def node_b_parent(state):
+        return {"dialog_state": ["node_b_parent"]}
+
+    main_builder = StateGraph(State)
+    main_builder.add_node(node_b_parent)
+    main_builder.add_edge(START, "subgraph_node")
+    main_builder.add_node("subgraph_node", sub_graph, destinations=("node_b_parent",))
+
+    main_graph = main_builder.compile(async_checkpointer, name="parent")
+    config = {"configurable": {"thread_id": 1}}
+
+    assert await main_graph.ainvoke(
+        input={"dialog_state": ["init_state"]}, config=config
+    ) == {"dialog_state": ["init_state", "b_child_state", "node_b_parent"]}

libs/langgraph/uv.lock

@@ -1197,7 +1197,7 @@ wheels = [
 
 [[package]]
 name = "langgraph"
-version = "0.4.5"
+version = "0.4.8"
 source = { editable = "." }
 dependencies = [
     { name = "langchain-core" },
@@ -1416,7 +1416,7 @@ inmem = [
 
 [[package]]
 name = "langgraph-prebuilt"
-version = "0.1.8"
+version = "0.2.2"
 source = { editable = "../prebuilt" }
 dependencies = [
     { name = "langchain-core" },

libs/prebuilt/langgraph/prebuilt/chat_agent_executor.py

@@ -240,7 +240,7 @@ def _validate_chat_history(
 
 def create_react_agent(
     model: Union[str, LanguageModelLike],
-    tools: Union[Sequence[Union[BaseTool, Callable]], ToolNode],
+    tools: Union[Sequence[Union[BaseTool, Callable, dict[str, Any]]], ToolNode],
     *,
     prompt: Optional[Prompt] = None,
     response_format: Optional[
@@ -420,12 +420,13 @@ def create_react_agent(
             else AgentState
         )
 
+    llm_builtin_tools: list[dict] = []
     if isinstance(tools, ToolNode):
         tool_classes = list(tools.tools_by_name.values())
         tool_node = tools
     else:
-        tool_node = ToolNode(tools)
-        # get the tool functions wrapped in a tool class from the ToolNode
+        llm_builtin_tools = [t for t in tools if isinstance(t, dict)]
+        tool_node = ToolNode([t for t in tools if not isinstance(t, dict)])
         tool_classes = list(tool_node.tools_by_name.values())
 
     if isinstance(model, str):
@@ -442,8 +443,12 @@ def create_react_agent(
 
     tool_calling_enabled = len(tool_classes) > 0
 
-    if _should_bind_tools(model, tool_classes) and tool_calling_enabled:
-        model = cast(BaseChatModel, model).bind_tools(tool_classes)
+    if (
+        _should_bind_tools(model, tool_classes)
+        and len(tool_classes) > 0
+        or (len(llm_builtin_tools) > 0)
+    ):
+        model = cast(BaseChatModel, model).bind_tools(tool_classes + llm_builtin_tools)  # type: ignore[operator]
 
     model_runnable = _get_prompt_runnable(prompt) | model
 
@@ -722,6 +727,10 @@ def create_react_agent(
             ]
 
             if pending_tool_calls:
+                pending_tool_calls = [
+                    tool_node.inject_tool_args(call, state, store)  # type: ignore[arg-type]
+                    for call in pending_tool_calls
+                ]
                 return [Send("tools", [tool_call]) for tool_call in pending_tool_calls]
             elif isinstance(messages[-1], ToolMessage):
                 return entrypoint

libs/prebuilt/langgraph/prebuilt/interrupt.py

@@ -1,12 +1,7 @@
-from copy import deepcopy
-from typing import Any, Literal, Optional, Union, cast
+from typing import Literal, Optional, Union
 
-from langchain_core.messages import ToolCall, ToolMessage
 from typing_extensions import TypedDict
 
-from langgraph.types import Command, interrupt
-from langgraph.utils.runnable import RunnableCallable
-
 
 class HumanInterruptConfig(TypedDict):
     """Configuration that defines what actions are allowed for a human interrupt.
@@ -93,159 +88,3 @@ class HumanResponse(TypedDict):
 
     type: Literal["accept", "ignore", "response", "edit"]
     args: Union[None, str, ActionRequest]
-
-
-class InterruptToolNode(RunnableCallable):
-    """Prebuilt post model hook node used to enable common patterns for tool interrupts.
-
-    For any tools with specified policies, an interrupt will be raised when the LLM returns
-    a tool call for said tool. The interrupt policy will be used to determine what sort of resume logic is allowed.
-    Any of the following resume patterns are supported:
-
-    * accept: the tool call is executed as planned
-    * edit: the args for the tool call are edited and then the tool call is executed
-    * response: text response/feedback is fed back into the LLM
-    * ignore: the current tool call is ignored / skipped
-
-    Args:
-        **interrupt_policy: a mapping of tool names to [`HumanInterruptConfig`][prebuilt.interrupt.HumanInterruptConfig] dictionaries
-            specifying which interrupt patterns to enable for said tool.
-
-        Example:
-        ```python
-        from langgraph.prebuilt import create_react_agent
-        from langgraph.checkpoint.memory import InMemorySaver
-        from langgraph.prebuilt.interrupt import HumanInterruptConfig, InterruptToolNode
-        from langgraph.types import Command
-
-
-        def book_hotel(hotel_name: str) -> str:
-            '''Book a room at the provided hotel.'''
-            # Some hotel API calls, a sensitive / expensive operation
-            return f"Booked a hotel at {hotel_name}."
-
-
-        agent = create_react_agent(
-            "openai:gpt-4.1",
-            tools=[book_hotel],
-            prompt="You are a hotel booking assistant.",
-            post_model_hook=InterruptToolNode(
-                book_hotel=HumanInterruptConfig(
-                    allow_accept=True,
-                    allow_edit=True,
-                    allow_ignore=True,
-                    allow_respond=True,
-                )
-            ),
-            checkpointer=InMemorySaver(),
-        )
-
-        config = {"configurable": {"thread_id": 1}}
-
-        response = agent.invoke(
-            {"messages": [{"role": "user", "content": "please book a hotel at the hilton inn in boston."}]},
-            config=config,
-        )
-
-        response = agent.invoke(Command(resume={"type": "accept"}), config=config)
-        ```
-    """
-
-    def __init__(self, **interrupt_policy: HumanInterruptConfig):
-        super().__init__(self._func, self._afunc)
-        self.interrupt_policy = interrupt_policy
-
-    def _interrupt(
-        self,
-        tool_call: ToolCall,
-        interrupt_config: HumanInterruptConfig,
-    ) -> Union[ToolCall, ToolMessage]:
-        """Interrupt before a tool call and ask for human input."""
-        call_id = tool_call["id"]
-        tool_name = tool_call["name"]
-
-        request = HumanInterrupt(
-            action_request=ActionRequest(
-                action=tool_name,
-                args=tool_call["args"],
-            ),
-            config=interrupt_config,
-            description=f"Please review tool call for `{tool_name}` before execution.",
-        )
-        response = interrupt([request])
-
-        # resume provided by agent inbox as a list
-        response = response[0] if isinstance(response, list) else response
-
-        try:
-            response_type = response.get("type")
-        except AttributeError:
-            raise TypeError(
-                f"Unexpected resume value: {response}."
-                f"Expected a dict with `'type'` key."
-            )
-
-        if response_type == "accept" and interrupt_config["allow_accept"]:
-            return tool_call
-        elif response_type == "edit" and interrupt_config["allow_edit"]:
-            return ToolCall(
-                args=cast(ActionRequest, response)["args"]["args"],
-                name=tool_name,
-                id=call_id,
-                type="tool_call",
-            )
-        elif response_type == "response" and interrupt_config["allow_respond"]:
-            return ToolMessage(
-                content=cast(str, response["args"]),
-                name=tool_name,
-                tool_call_id=call_id,
-                status="error",
-            )
-        elif response_type == "ignore" and interrupt_config["allow_ignore"]:
-            return ToolMessage(
-                content=f"User ignored the tool call for `{tool_name}` with id {call_id}",
-                name=tool_name,
-                tool_call_id=call_id,
-                status="success",
-            )
-
-        allowed_types = [
-            type_name
-            for type_name, is_allowed in {
-                "accept": interrupt_config["allow_accept"],
-                "edit": interrupt_config["allow_edit"],
-                "response": interrupt_config["allow_respond"],
-                "ignore": interrupt_config["allow_ignore"],
-            }.items()
-            if is_allowed
-        ]
-
-        raise ValueError(
-            f"Unexpected human response: {response}. "
-            f"Expected one with `'type'` in {allowed_types} based on {tool_name}'s interrupt configuration."
-        )
-
-    def _func(self, input: dict[str, Any]) -> Command:
-        ai_msg = input["messages"][-1]
-        tool_calls: list[ToolCall] = deepcopy(ai_msg.tool_calls) or []
-        tool_messages: list[ToolMessage] = []
-
-        for idx, tool_call in enumerate(tool_calls):
-            if interrupt_config := self.interrupt_policy.get(tool_call["name"]):
-                interrupt_result = self._interrupt(
-                    tool_call=tool_call, interrupt_config=interrupt_config
-                )
-
-                if isinstance(interrupt_result, ToolMessage):
-                    tool_messages.append(interrupt_result)
-                else:
-                    tool_calls[idx] = interrupt_result
-
-        updated_ai_msg = ai_msg.copy(update={"tool_calls": tool_calls})
-
-        # conditional routing logic for post_model_hook will direct to the tools node
-        # or agent node depending on if there are pending tool calls
-        return {"messages": [updated_ai_msg, *tool_messages]}
-
-    async def _afunc(self, input: dict[str, Any]) -> Command:
-        return self._func(input)

libs/prebuilt/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "langgraph-prebuilt"
-version = "0.1.8"
+version = "0.2.2"
 description = "Library with high-level APIs for creating and executing LangGraph agents and tools."
 authors = []
 requires-python = ">=3.9"

libs/prebuilt/tests/test_interrupt_tool_node.py

@@ -1,191 +0,0 @@
-import pytest
-from langchain_core.messages import ToolMessage
-from langchain_core.runnables import RunnableConfig
-
-from langgraph.checkpoint.base import BaseCheckpointSaver
-from langgraph.prebuilt import create_react_agent
-from langgraph.prebuilt.interrupt import HumanInterruptConfig, InterruptToolNode
-from langgraph.types import Command
-from tests.model import FakeToolCallingModel
-
-
-def hello_tool(name: str) -> str:
-    """Return a greeting for the provided person."""
-    return f"Hello, {name}!"
-
-
-post_model_hook = InterruptToolNode(
-    hello_tool=HumanInterruptConfig(
-        allow_accept=True,
-        allow_edit=True,
-        allow_ignore=True,
-        allow_respond=True,
-    )
-)
-
-default_model = FakeToolCallingModel(
-    tool_calls=[
-        [
-            {
-                "name": "hello_tool",
-                "args": {"name": "lady gaga"},
-                "id": "some-random-id",
-            }
-        ]
-    ]
-)
-
-
-def test_interrupt_surfaced(
-    request: pytest.FixtureRequest,
-    sync_checkpointer: BaseCheckpointSaver,
-) -> None:
-    agent = create_react_agent(
-        default_model,
-        [hello_tool],
-        checkpointer=sync_checkpointer,
-        post_model_hook=post_model_hook,
-    )
-    config: RunnableConfig = {"configurable": {"thread_id": "1"}}
-    result = agent.invoke({"messages": [("user", "Say hi to lady gaga!")]}, config)
-
-    interrupt_data = result["__interrupt__"]
-    assert interrupt_data[0].value == [
-        {
-            "action_request": {"action": "hello_tool", "args": {"name": "lady gaga"}},
-            "config": {
-                "allow_accept": True,
-                "allow_edit": True,
-                "allow_ignore": True,
-                "allow_respond": True,
-            },
-            "description": "Please review tool call for `hello_tool` before execution.",
-        }
-    ]
-
-    response = agent.invoke(Command(resume={"type": "accept"}), config=config)
-    tool_message: ToolMessage = response["messages"][-2]
-    assert tool_message.content == "Hello, lady gaga!"
-    assert tool_message.name == "hello_tool"
-
-
-@pytest.mark.parametrize(
-    "resume, expected_content",
-    [
-        ({"type": "accept"}, "Hello, lady gaga!"),
-        (
-            {"type": "ignore"},
-            "User ignored the tool call for `hello_tool` with id some-random-id",
-        ),
-        (
-            {
-                "type": "edit",
-                "args": {"action": "hello_tool", "args": {"name": "bruno mars"}},
-            },
-            "Hello, bruno mars!",
-        ),
-    ],
-)
-def test_interrupt_resume_variants(
-    request: pytest.FixtureRequest,
-    sync_checkpointer: BaseCheckpointSaver,
-    resume: dict,
-    expected_content: str,
-) -> None:
-    agent = create_react_agent(
-        default_model,
-        [hello_tool],
-        checkpointer=sync_checkpointer,
-        post_model_hook=post_model_hook,
-    )
-
-    config: RunnableConfig = {"configurable": {"thread_id": "1"}}
-    agent.invoke({"messages": [("user", "Say hi to lady gaga!")]}, config)
-
-    response = agent.invoke(Command(resume=resume), config=config)
-    tool_message: ToolMessage = response["messages"][-2]
-    assert tool_message.name == "hello_tool"
-    assert tool_message.content == expected_content
-
-    if resume["type"] == "edit":
-        ai_msg = response["messages"][-1]
-        assert ai_msg.tool_calls == [
-            {
-                "name": "hello_tool",
-                "args": {"name": "lady gaga"},
-                "id": "some-random-id",
-                "type": "tool_call",
-            }
-        ]
-
-
-def test_resume_with_response(
-    request: pytest.FixtureRequest,
-    sync_checkpointer: BaseCheckpointSaver,
-) -> None:
-    model = FakeToolCallingModel(
-        tool_calls=[
-            [
-                {
-                    "name": "hello_tool",
-                    "args": {"name": "lady gaga"},
-                    "id": "some-random-id",
-                }
-            ],
-            [
-                {
-                    "name": "hello_tool",
-                    "args": {"name": "bruno mars"},
-                    "id": "some-random-id-2",
-                }
-            ],
-        ]
-    )
-
-    agent = create_react_agent(
-        model,
-        [hello_tool],
-        checkpointer=sync_checkpointer,
-        post_model_hook=post_model_hook,
-    )
-
-    config: RunnableConfig = {"configurable": {"thread_id": "1"}}
-    agent.invoke({"messages": [("user", "Say hi to lady gaga!")]}, config)
-
-    # Provide user response
-    agent.invoke(
-        Command(
-            resume={
-                "type": "response",
-                "args": "actually, please say hello to bruno mars",
-            }
-        ),
-        config=config,
-    )
-
-    # Accept the updated call
-    response = agent.invoke(Command(resume={"type": "accept"}), config=config)
-
-    assert len(response["messages"]) == 6
-    tool_message: ToolMessage = response["messages"][-2]
-    assert tool_message.name == "hello_tool"
-    assert tool_message.content == "Hello, bruno mars!"
-
-
-def test_resume_with_type_not_allowed(sync_checkpointer: BaseCheckpointSaver) -> None:
-    agent = create_react_agent(
-        default_model,
-        [hello_tool],
-        checkpointer=sync_checkpointer,
-        post_model_hook=post_model_hook,
-    )
-    config: RunnableConfig = {"configurable": {"thread_id": "1"}}
-    agent.invoke({"messages": [("user", "Say hi to lady gaga!")]}, config)
-
-    with pytest.raises(ValueError) as exc_info:
-        agent.invoke(Command(resume={"type": "not-allowed"}), config=config)
-
-    assert (
-        str(exc_info.value)
-        == "Unexpected human response: {'type': 'not-allowed'}. Expected one with `'type'` in ['accept', 'edit', 'response', 'ignore'] based on hello_tool's interrupt configuration."
-    )

libs/prebuilt/tests/test_react_agent.py

@@ -1540,3 +1540,63 @@ def test_post_model_hook_with_structured_output() -> None:
             }
         },
     ]
+
+
+@pytest.mark.parametrize(
+    "state_schema", [AgentStateExtraKey, AgentStateExtraKeyPydantic]
+)
+def test_create_react_agent_inject_vars_with_post_model_hook(
+    state_schema: StateSchemaType,
+) -> None:
+    store = InMemoryStore()
+    namespace = ("test",)
+    store.put(namespace, "test_key", {"bar": 3})
+
+    if issubclass(state_schema, AgentStatePydantic):
+
+        def tool1(
+            some_val: int,
+            state: Annotated[AgentStateExtraKeyPydantic, InjectedState],
+            store: Annotated[BaseStore, InjectedStore()],
+        ) -> str:
+            """Tool 1 docstring."""
+            store_val = store.get(namespace, "test_key").value["bar"]
+            return some_val + state.foo + store_val
+    else:
+
+        def tool1(
+            some_val: int,
+            state: Annotated[dict, InjectedState],
+            store: Annotated[BaseStore, InjectedStore()],
+        ) -> str:
+            """Tool 1 docstring."""
+            store_val = store.get(namespace, "test_key").value["bar"]
+            return some_val + state["foo"] + store_val
+
+    tool_call = {
+        "name": "tool1",
+        "args": {"some_val": 1},
+        "id": "some 0",
+        "type": "tool_call",
+    }
+
+    def post_model_hook(state: dict) -> None:
+        return
+
+    model = FakeToolCallingModel(tool_calls=[[tool_call], []])
+    agent = create_react_agent(
+        model,
+        [tool1],
+        state_schema=state_schema,
+        store=store,
+        post_model_hook=post_model_hook,
+    )
+    input_message = HumanMessage("hi")
+    result = agent.invoke({"messages": [input_message], "foo": 2})
+    assert result["messages"] == [
+        input_message,
+        AIMessage(content="hi", tool_calls=[tool_call], id="0"),
+        _AnyIdToolMessage(content="6", name="tool1", tool_call_id="some 0"),
+        AIMessage("hi-hi-6", id="1"),
+    ]
+    assert result["foo"] == 2

libs/sdk-js/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@langchain/langgraph-sdk",
-  "version": "0.0.77",
+  "version": "0.0.78",
   "description": "Client library for interacting with the LangGraph API",
   "type": "module",
   "packageManager": "yarn@1.22.19",

libs/sdk-js/src/react/stream.tsx

@@ -457,6 +457,11 @@ export interface UseStreamOptions<
    */
   onFinish?: (state: ThreadState<StateType>) => void;
 
+  /**
+   * Callback that is called when a new stream is created.
+   */
+  onCreated?: (run: { run_id: string; thread_id: string }) => void;
+
   /**
    * Callback that is called when an update event is received.
    */
@@ -502,6 +507,15 @@ export interface UseStreamOptions<
    * Callback that is called when the thread ID is updated (ie when a new thread is created).
    */
   onThreadId?: (threadId: string) => void;
+
+  /** Will reconnect the stream on mount */
+  reconnectOnMount?: boolean | (() => RunMetadataStorage);
+}
+
+interface RunMetadataStorage {
+  getItem(key: `lg:stream:${string}`): string | null;
+  setItem(key: `lg:stream:${string}`, value: string): void;
+  removeItem(key: `lg:stream:${string}`): void;
 }
 
 export interface UseStream<
@@ -590,6 +604,11 @@ export interface UseStream<
    * The ID of the assistant to use.
    */
   assistantId: string;
+
+  /**
+   * Join an active stream.
+   */
+  joinStream: (runId: string) => Promise<void>;
 }
 
 type ConfigWithConfigurable<ConfigurableType extends Record<string, unknown>> =
@@ -619,6 +638,7 @@ interface SubmitOptions<
    * @default false
    */
   streamSubgraphs?: boolean;
+  streamResumable?: boolean;
 }
 
 export function useStream<
@@ -647,7 +667,17 @@ export function useStream<
     | ErrorStreamEvent
     | FeedbackStreamEvent;
 
-  let { assistantId, messagesKey, onError, onFinish } = options;
+  let { assistantId, messagesKey, onCreated, onError, onFinish } = options;
+
+  const reconnectOnMountRef = useRef(options.reconnectOnMount);
+  const runMetadataStorage = useMemo(() => {
+    if (typeof window === "undefined") return null;
+    const storage = reconnectOnMountRef.current;
+    if (storage === true) return window.sessionStorage;
+    if (typeof storage === "function") return storage();
+    return null;
+  }, []);
+
   messagesKey ??= "messages";
 
   const client = useMemo(
@@ -722,6 +752,7 @@ export function useStream<
 
   // TODO: this should be done on the server to avoid pagination
   // TODO: should we permit adapter? SWR / React Query?
+  // TODO: make this only when branching is expected
   const history = useThreadHistory<StateType>(
     threadId,
     client,
@@ -800,15 +831,23 @@ export function useStream<
     );
   })();
 
-  const stop = useCallback(() => {
+  const stop = () => {
     if (abortRef.current != null) abortRef.current.abort();
     abortRef.current = null;
-  }, []);
 
-  const submit = async (
-    values: UpdateType | null | undefined,
-    submitOptions?: SubmitOptions<StateType, ConfigurableType>,
-  ) => {
+    if (runMetadataStorage && threadId) {
+      const runId = runMetadataStorage.getItem(`lg:stream:${threadId}`);
+      if (runId) client.runs.cancel(threadId, runId);
+      runMetadataStorage.removeItem(`lg:stream:${threadId}`);
+    }
+  };
+
+  async function consumeStream(
+    action: (signal: AbortSignal) => Promise<{
+      onSuccess: () => Promise<ThreadState<StateType>[]>;
+      stream: AsyncGenerator<EventStreamEvent>;
+    }>,
+  ) {
     try {
       setIsLoading(true);
       setStreamError(undefined);
@@ -816,69 +855,10 @@ export function useStream<
       submittingRef.current = true;
       abortRef.current = new AbortController();
 
-      // Unbranch things
-      const newPath = submitOptions?.checkpoint?.checkpoint_id
-        ? branchByCheckpoint[submitOptions?.checkpoint?.checkpoint_id]?.branch
-        : undefined;
-
-      if (newPath != null) setBranch(newPath ?? "");
-
-      // Assumption: we're setting the initial value
-      // Used for instant feedback
-      setStreamValues(() => {
-        const values = { ...historyValues };
-
-        if (submitOptions?.optimisticValues != null) {
-          return {
-            ...values,
-            ...(typeof submitOptions.optimisticValues === "function"
-              ? submitOptions.optimisticValues(values)
-              : submitOptions.optimisticValues),
-          };
-        }
-
-        return values;
-      });
-
-      let usableThreadId = threadId;
-      if (!usableThreadId) {
-        const thread = await client.threads.create();
-        onThreadId(thread.thread_id);
-        usableThreadId = thread.thread_id;
-      }
-
-      const streamMode = unique([
-        ...(submitOptions?.streamMode ?? []),
-        ...trackStreamModeRef.current,
-        ...callbackStreamMode,
-      ]);
-
-      const checkpoint =
-        submitOptions?.checkpoint ?? threadHead?.checkpoint ?? undefined;
-      // @ts-expect-error
-      if (checkpoint != null) delete checkpoint.thread_id;
-
-      const run = client.runs.stream(usableThreadId, assistantId, {
-        input: values as Record<string, unknown>,
-        config: submitOptions?.config,
-        command: submitOptions?.command,
-
-        interruptBefore: submitOptions?.interruptBefore,
-        interruptAfter: submitOptions?.interruptAfter,
-        metadata: submitOptions?.metadata,
-        multitaskStrategy: submitOptions?.multitaskStrategy,
-        onCompletion: submitOptions?.onCompletion,
-        onDisconnect: submitOptions?.onDisconnect ?? "cancel",
-
-        signal: abortRef.current.signal,
-
-        checkpoint,
-        streamMode,
-        streamSubgraphs: submitOptions?.streamSubgraphs,
-      }) as AsyncGenerator<EventStreamEvent>;
+      const run = await action(abortRef.current.signal);
 
       let streamError: StreamError | undefined;
-      for await (const { event, data } of run) {
+      for await (const { event, data } of run.stream) {
         if (event === "error") {
           streamError = new StreamError(data);
           break;
@@ -930,9 +910,9 @@ export function useStream<
       }
 
       // TODO: stream created checkpoints to avoid an unnecessary network request
-      const result = await history.mutate(usableThreadId);
-      setStreamValues(null);
+      const result = await run.onSuccess();
 
+      setStreamValues(null);
       if (streamError != null) throw streamError;
 
       const lastHead = result.at(0);
@@ -956,8 +936,146 @@ export function useStream<
       submittingRef.current = false;
       abortRef.current = null;
     }
+  }
+
+  const joinStream = async (runId: string, lastEventId?: string) => {
+    lastEventId ??= "-1";
+    if (!threadId) return;
+    await consumeStream(async (signal: AbortSignal) => {
+      const stream = client.runs.joinStream(threadId, runId, {
+        signal,
+        lastEventId,
+      }) as AsyncGenerator<EventStreamEvent>;
+
+      return {
+        onSuccess: () => {
+          runMetadataStorage?.removeItem(`lg:stream:${threadId}`);
+          return history.mutate(threadId);
+        },
+        stream,
+      };
+    });
   };
 
+  const submit = async (
+    values: UpdateType | null | undefined,
+    submitOptions?: SubmitOptions<StateType, ConfigurableType>,
+  ) => {
+    await consumeStream(async (signal: AbortSignal) => {
+      // Unbranch things
+      const newPath = submitOptions?.checkpoint?.checkpoint_id
+        ? branchByCheckpoint[submitOptions?.checkpoint?.checkpoint_id]?.branch
+        : undefined;
+
+      if (newPath != null) setBranch(newPath ?? "");
+
+      // Assumption: we're setting the initial value
+      // Used for instant feedback
+      setStreamValues(() => {
+        const values = { ...historyValues };
+
+        if (submitOptions?.optimisticValues != null) {
+          return {
+            ...values,
+            ...(typeof submitOptions.optimisticValues === "function"
+              ? submitOptions.optimisticValues(values)
+              : submitOptions.optimisticValues),
+          };
+        }
+
+        return values;
+      });
+
+      let usableThreadId = threadId;
+      if (!usableThreadId) {
+        const thread = await client.threads.create();
+        onThreadId(thread.thread_id);
+        usableThreadId = thread.thread_id;
+      }
+
+      const streamMode = unique([
+        ...(submitOptions?.streamMode ?? []),
+        ...trackStreamModeRef.current,
+        ...callbackStreamMode,
+      ]);
+
+      const checkpoint =
+        submitOptions?.checkpoint ?? threadHead?.checkpoint ?? undefined;
+      // @ts-expect-error
+      if (checkpoint != null) delete checkpoint.thread_id;
+      let rejoinKey: `lg:stream:${string}` | undefined;
+
+      const stream = client.runs.stream(usableThreadId, assistantId, {
+        input: values as Record<string, unknown>,
+        config: submitOptions?.config,
+        command: submitOptions?.command,
+
+        interruptBefore: submitOptions?.interruptBefore,
+        interruptAfter: submitOptions?.interruptAfter,
+        metadata: submitOptions?.metadata,
+        multitaskStrategy: submitOptions?.multitaskStrategy,
+        onCompletion: submitOptions?.onCompletion,
+        onDisconnect:
+          submitOptions?.onDisconnect ??
+          (runMetadataStorage ? "continue" : "cancel"),
+
+        signal,
+
+        checkpoint,
+        streamMode,
+        streamSubgraphs: submitOptions?.streamSubgraphs,
+        streamResumable: submitOptions?.streamResumable ?? !!runMetadataStorage,
+        onRunCreated(params) {
+          const runParams = {
+            run_id: params.run_id,
+            thread_id: params.thread_id ?? usableThreadId,
+          };
+          if (runMetadataStorage) {
+            rejoinKey = `lg:stream:${runParams.thread_id}`;
+            runMetadataStorage.setItem(rejoinKey, runParams.run_id);
+          }
+          onCreated?.(runParams);
+        },
+      }) as AsyncGenerator<EventStreamEvent>;
+
+      return {
+        stream,
+        onSuccess: () => {
+          if (rejoinKey) runMetadataStorage?.removeItem(rejoinKey);
+          return history.mutate(usableThreadId);
+        },
+      };
+    });
+  };
+
+  const reconnectKey = useMemo(() => {
+    if (!runMetadataStorage || isLoading) return undefined;
+    if (typeof window === "undefined") return undefined;
+    const runId = runMetadataStorage?.getItem(`lg:stream:${threadId}`);
+    if (!runId) return undefined;
+    return { runId, threadId };
+  }, [runMetadataStorage, isLoading, threadId]);
+
+  const shouldReconnect = !!runMetadataStorage;
+  const reconnectRef = useRef({ threadId, shouldReconnect });
+
+  const joinStreamRef = useRef<typeof joinStream>(joinStream);
+  joinStreamRef.current = joinStream;
+
+  useEffect(() => {
+    // reset shouldReconnect when switching threads
+    if (reconnectRef.current.threadId !== threadId) {
+      reconnectRef.current = { threadId, shouldReconnect };
+    }
+  }, [threadId, shouldReconnect]);
+
+  useEffect(() => {
+    if (reconnectKey && reconnectRef.current.shouldReconnect) {
+      reconnectRef.current.shouldReconnect = false;
+      joinStreamRef.current?.(reconnectKey.runId);
+    }
+  }, [reconnectKey]);
+
   const error = streamError ?? historyError;
   const values = streamValues ?? historyValues;
 
@@ -976,6 +1094,8 @@ export function useStream<
     stop,
     submit,
 
+    joinStream,
+
     branch,
     setBranch,