LangGraph
AI & LLMsBuild resilient language agents as graphs.
Release History
checkpoint==4.0.11 featureThis release introduces the addition of serde events and bumps the version to 0.4.1.
1.0.10This release primarily consists of internal maintenance, dependency updates, and version bumps for langgraph and checkpoint sub-packages.
checkpoint==4.0.1rc3This release primarily consists of chore updates, including updating default settings and adding support for workflow dispatch in CI.
checkpoint==4.0.1rc2This release is a release candidate (rc2) and includes an improvement to subclass handling.
checkpoint==4.0.1rc1This release primarily focuses on dependency updates, conformance testing, and minor documentation clarity improvements.
1.0.10rc11 fixThis release focuses on internal improvements, adding type checking targets, and fixing a specific bubbling issue within langgraph related to checkpoint namespaces.
sdk==0.3.91 featureThis release introduces a new 'extract' parameter for the threads.search() method and adds a type checking target to the build process.
sdk==0.3.82 featuresThis release introduces new streaming and durability options for crons and enhances type safety for store authentication within the SDK.
1.0.91 fix1 featureThis release introduces langgraph prebuilt components and fixes an issue related to sequential interrupt handling in the functional API. It also includes several dependency bumps and minor internal refactoring.
prebuilt==1.0.81 fix1 featureThis release introduces integration with langgraph and fixes an issue related to dynamically registered tool injection. It also includes several dependency updates.
sdk==0.3.71 fixThis release focuses on minor fixes, including allowing configuration resets during assistant updates, and dependency bumps for internal libraries.
sdk==0.3.61 featureThis release introduces a new prune method and includes internal reorganization of client files.
sdk==0.3.5This release includes a chore update related to the server runtime type.
1.0.81 fixThis release includes a fix for double streaming in pydantic messages and various documentation enhancements across the SDK and CLI.
sdk==0.3.42 featuresThis release introduces new functionality for managing crons, including an update method and enabling/disabling capabilities, alongside dependency updates.
checkpointpostgres==3.0.4This release primarily contains maintenance updates, including dependency upgrades and minor chore changes like omitting the lock file when using a connection pool and updating Twitter URLs.
cli==0.4.12This release primarily updates dependencies and corrects outdated Twitter URLs within the CLI tool.
1.0.7Breaking1 fixVersion 1.0.7 includes dependency upgrades and a critical fix addressing a breaking change introduced in aiosqlite. Twitter URLs were also updated.
prebuilt==1.0.7Breaking1 fix1 featureThis release updates the prebuilt package to version 1.0.7, introducing dynamic tool calling support and fixing a bug related to aiosqlite.
checkpointsqlite==3.0.31 fixThis release primarily addresses a breaking change introduced by aiosqlite and includes dependency upgrades.
sdk==0.3.31 featureThis release improves error messages and introduces end-time functionality to the crons client.
checkpointpostgres==3.0.31 fixThis release primarily addresses a bug fix by flipping the default setting on the base cache and includes minor documentation improvements.
1.0.64 fixes1 featureThis release includes several bug fixes, such as correcting a typo and sanitizing namespaces, alongside a new feature to validate checkpointer types at compile time.
prebuilt==1.0.65 fixes3 featuresThis release includes several bug fixes, such as addressing issues with cache defaults and tool runtime injection, alongside new features like custom encryption at rest and pagination support in assistant search responses.
checkpointsqlite==3.0.21 fixThis release primarily addresses a bug fix related to the base cache default setting and includes minor documentation updates.
checkpoint==4.0.03 fixesThis release focuses on minor fixes, including correcting the default setting for the base cache and ensuring InMemorySaver context managers behave correctly.
sdk==0.3.2sdk==0.3.1Breaking2 fixesVersion 0.3.1 removes model-type specific custom JSON encryption annotations and adds documentation regarding key preservation limitations.
cli==0.4.111 featureMinor update to the CLI to relax API version constraints.
cli==0.4.101 fixThis patch release fixes a bug in the CLI's compose code generation by properly escaping variable substitutions.
1.0.52 fixes3 featuresThis release introduces custom encryption at rest, pagination for assistant searches, and includes ID fields in Python SDK stream events, alongside version bumps for checkpointing packages.
sdk==0.3.01 featureThis release introduces the inclusion of event IDs within stream events for the Python SDK.
checkpointpostgres==3.0.22 fixesThis release focuses on stability and security improvements for the PostgreSQL checkpoint saver, specifically addressing SQL parameterization in migrations and idempotent extension creation.
checkpointsqlite==3.0.11 fixThis release focuses on hardening the checkpoint-sqlite implementation, updating ormsgpack constraints, and adding project metadata and license files.
sdk==0.2.151 featureThis release introduces a new filtering capability to the Assistants count API by adding a name parameter.
cli==0.4.91 featureThis release updates the CLI dev server to support webhook configuration pass-through.
cli==0.4.82 featuresThis release introduces webhook configuration and custom encryption at rest for the CLI, alongside schema updates for langgraph.json.
sdk==0.2.142 featuresThis release introduces custom encryption at rest and configures the loopback client for the Python SDK.
sdk==0.2.122 featuresThis release introduces pagination for assistant search results and adds a mechanism to bypass automatic API key loading. It also includes improvements to input type-hinting across the Python SDK.
1.0.43 fixes2 featuresVersion 1.0.4 focuses on stability improvements, including fixes for interrupt stream modes, enhanced type checking in the SDK, and the removal of thread ID from configurable fields in remote graphs.
sdk==0.2.10Breaking4 fixes4 featuresRelease 0.2.10 adds a name parameter to the Assistants search API, introduces Python 3.14 support, and drops support for Python 3.9 while improving type safety and documentation.
prebuilt==1.0.51 fixRelease 1.0.5 focuses on internal refactoring of injection logic to respect function signatures and deprecates the use of setattr on ToolCallRequest.
prebuilt==1.0.41 fixThis patch release addresses dependency warnings within the prebuilt module to improve console output and stability.
prebuilt==1.0.32 fixes2 featuresRelease 1.0.3 focuses on dependency maintenance, including a bump for the Postgres checkpointer and improved surrogate handling in ormsgpack.
1.0.36 fixes2 featuresThis patch release focuses on documentation synchronization, linting fixes for REMOVE_ALL_MESSAGES, and critical bug fixes regarding state updates and stream_mode sequences.
checkpointpostgres==3.0.11 fixThis release fixes an idempotency issue in async Postgres migrations and updates the ormsgpack dependency with new serialization options.
checkpoint==3.0.11 fix1 featureThis release updates the ormsgpack dependency constraints, adds a new serialization option, and corrects the checkpoint interface documentation.
cli==0.4.7No release notes provided.
cli==0.4.61 fixThis release includes a fix for buildkit syntax directives and updates to the CLI configuration schema.
cli==0.4.51 featureThis release adds cursory support for Python 3.14.
1.0.2Breaking3 fixes4 featuresThis release marks the major v1.0.0 milestone for LangGraph, dropping Python 3.9 support and introducing Checkpointers 3.0 along with the new Overwrite reducer bypass.
prebuilt==1.0.2Breaking2 fixes4 featuresThis release marks the major v1.0.0 milestone for langgraph and langgraph-prebuilt, dropping support for Python 3.9 and introducing Checkpointers 3.0.
prebuilt==0.6.51 fix3 featuresThis release includes the implementation of Redis node-level caching, support for client query parameters, and various dependency bumps including support for checkpoint 3.0.
0.6.11This maintenance release updates dependency constraints to allow the use of checkpoint 3.0.
1.0.1Breaking1 featureThis release upgrades Checkpointers to version 3.0 and implements stricter security restrictions on JSON type deserialization.
prebuilt==1.0.11 featureThis release introduces version 3.0 of the Checkpointers module.
checkpointpostgres==3.0.0Breaking2 featuresThis major release upgrades checkpointers to version 3.0 and drops support for Python 3.9 while introducing preliminary support for Python 3.14.
checkpointsqlite==3.0.0Breaking2 fixes3 featuresThis major release upgrades checkpointers to version 3.0, drops support for Python 3.9, and introduces Redis node-level caching alongside various bug fixes for SQLite and memory stores.
checkpoint==3.0.0Breaking2 fixes3 featuresThis release marks the transition to Checkpointers 3.0 and officially drops support for Python 3.9 while adding preliminary support for Python 3.14.
1.0.01 fix2 featuresThis release marks the official 1.0.0 version of langgraph and langgraph-prebuilt, including preliminary support for Python 3.14 and documentation refinements.
prebuilt==1.0.01 fix2 featuresThis release marks the official v1.0.0 for langgraph and langgraph-prebuilt, including preliminary support for Python 3.14 and documentation refinements.
1.0.0rc1Breaking1 fix2 featuresThis release marks the transition to LangGraph v1 and officially drops support for Python 3.9. It includes various chores for prebuilt components and internal renaming.
prebuilt==0.7.0rc1Breaking3 featuresThis release marks the transition to Python 3.10+, un-deprecates ToolNode, and includes various version bumps for LangGraph core and its checkpointing extensions.
cli==0.4.4Breaking3 fixes1 featureThis release drops support for Python 3.9, renames 'studio' to 'debugger', and ensures local dependencies are installed in editable mode.
0.6.101 fixThis release primarily focuses on a bug fix for langgraph by reverting selective interrupt task scheduling and updating the langgraph dependency version.
cli==0.4.32 fixes3 featuresVersion 0.4.3 introduces authentication flags for custom routes, server customization ordering, and a new configuration schema, alongside improvements to version parsing and the context API.
0.6.95 fixesThis release focuses on bug fixes for task scheduling, state consistency, and result population, while reverting a caching behavior in nested resumes to ensure execution reliability.
checkpointpostgres==2.0.251 fixThis patch release fixes a serialization bug in checkpointpostgres related to handling metadata writes in older checkpoints when using Jsonb.
checkpoint==2.1.2Breaking5 fixes3 featuresThis release removes the legacy thread_ts alias, introduces Redis node-level caching, and improves checkpoint serialization and task cancellation resiliency.
checkpointpostgres==2.0.24Breaking2 fixes3 featuresThis release introduces a redis node level cache and durability mode while removing the deprecated thread_ts alias. It also includes critical fixes for Postgres checkpoint loading and non-ascii text preservation.
1.0.0a4No release notes provided.
0.6.8Breaking9 fixes4 featuresThis release focuses on stability and correctness in LangGraph, specifically addressing state management during resumes, graph rendering, and resource cleanup in asynchronous loops.
sdk==0.2.91 featureThis release improves the reliability of long-lived requests in the Python SDK by implementing reconnection logic for wait, join, and cancel endpoints.
sdk==0.2.81 fixThis release focuses on a bug fix for the Python SDK to properly handle Server-Sent Events (SSE) stream reconnections.
sdk==0.2.71 fixThis release fixes an issue with SSE stream flushing and includes various documentation cleanup and dependency updates.
1.0.0a3No release notes provided.
0.6.71 fix1 featureThis release introduces monorepo support for the CLI and fixes a type resolution bug involving Required/NotRequired annotations.
cli==0.4.21 featureThis release introduces support for prereleases within the CLI and includes a version bump.
cli==0.4.12 featuresThis release introduces preliminary support for monorepos and updates the configuration schema for LGP.
sdk==0.2.61 featureThis release introduces the ability to manage thread TTL (Time-to-Live) during creation and update operations in the Python SDK.
sdk==0.2.51 featureThis release introduces the ability to filter thread searches by specific IDs in the Python SDK.
1.0.0a2No release notes provided.
prebuilt==0.7.0a2This release marks the transition from version 0.7.0a1, though no specific changes were detailed in the provided notes.
sdk==0.2.42 featuresThis release introduces thread streaming capabilities and a new durability flag to the Python SDK.
1.0.0a1No release notes provided.
prebuilt==0.7.0a11 fix2 featuresThis release introduces Redis node-level caching and client query parameter support, alongside a fix for Remote Baggage in LangGraph.
cli==0.4.0Minor version update for the cli package from 0.3.8 to 0.4.0.
sdk==0.2.31 featureThis release introduces a new feature to the Python SDK allowing users to count endpoints.
cli==0.3.8This release updates the dependency constraints for the CLI, specifically bumping the maximum allowed version of langgraph-api.
cli==0.3.71 featureThis release updates the CLI to support newer Debian distributions including bookworm and trixie.
0.6.61 fix2 featuresThis release introduces passthrough parameters and headers for LangGraph methods and adds client query parameter support to the Python SDK, alongside a fix for Remote Baggage.
sdk==0.2.21 featureThis release introduces an asynchronous close method to the LangGraphClient and includes internal type updates for the Python SDK.
sdk==0.2.12 featuresThis release introduces support for client-side query parameters and a new Select-statement feature for the Python SDK.
0.6.51 fix1 featureThis release introduces Redis-based node-level caching for LangGraph and fixes a persistence bug related to resume_map values.
0.6.41 fix2 featuresThis release introduces performance optimizations for checkpointing and deprecates MessageGraph, while resolving mypy issues and improving subgraph message emission.
prebuilt==0.6.4Breaking2 fixesThis release deprecates MessageGraph, fixes message emission issues in subgraphs, and reverts a recent refactor to tool argument injection.
0.6.35 fixes1 featureThis release introduces a new durability mode for invoke/ainvoke and focuses on internal cleanup, including AgentState refinements and improved validation for node signatures.
prebuilt==0.6.31 fix1 featureThis release focuses on tidying up AgentState and improving type support for the model parameter in the create_react_agent function.
0.6.21 fix1 featureThis release introduces context coercion for the LangGraph runtime and fixes a schema assignment bug in prebuilt components.
prebuilt==0.6.21 fixVersion 0.6.2 of prebuilt includes a bug fix for the conditional assignment of context_schema to config_schema.
Common Errors
ModuleNotFoundError7 reportsThe "ModuleNotFoundError" in LangGraph usually indicates that required dependencies or LangGraph submodules are missing or not correctly installed. To fix this, ensure you've installed LangGraph and its dependencies using `pip install langgraph` (or `pip install "langgraph[all]"` for optional dependencies); try upgrading LangGraph to the latest version (`pip install --upgrade langgraph`) to resolve potential version-related issues. If the problem persists, double-check your Python environment and virtual environment activation.
BadRequestError6 reportsBadRequestError in langgraph usually arises from malformed requests to an external service, often an LLM, due to incorrect parameters or data. To fix it, carefully inspect the LLM's API call for configuration errors like conflicting parameters (e.g., both `max_tokens` and `max_completion_tokens`), invalid data types, or exceeding rate limits; correct any discrepancies or adjust input data to match the API's expected format. Ensure also that your loop termination condition is correctly defined in the graph.
ToolInvocationError2 reportsToolInvocationError in LangGraph usually arises from unhandled exceptions within a tool's execution or from Pydantic validation failures when passing arguments to tools, especially those with `ToolRuntime` parameters. To fix it, ensure your tool functions have robust error handling (try-except blocks) and meticulously validate the data passed to the tool using Pydantic schemas matching the expected input types, paying special attention to runtime dependencies. This ensures the tool executes properly and returns data in the expected format.
GraphLoadError2 reportsGraphLoadError in LangGraph often arises from missing or incorrect Python dependencies required by your graph's nodes and edges. Ensure all necessary packages, including those specified in `requirements.txt` or within your code (e.g., imports), are correctly installed using `pip install -r requirements.txt` (or equivalent) before running LangGraph. Consider utilizing a virtual environment to manage your project's dependencies and avoid conflicts with system-level packages. Double-check package names and versions for typos and compatibility.
NotImplementedError2 reportsThe "NotImplementedError" in LangGraph commonly arises when a required method or function within a custom class or component (like a custom Tool or Agent) is not defined. To fix this, identify the missing method (usually indicated in the error traceback) and implement its functionality within your custom class. Ensure your custom components align with the expected interface for proper execution within the LangGraph framework.
TypeError2 reportsThis TypeError usually arises when your LangGraph configuration, especially within the node context or checkpointing mechanism, attempts to serialize Python objects that aren't natively JSON serializable (e.g., custom classes, database connections, or complex objects). To fix it, either convert these problematic objects into JSON-serializable formats (like strings, dictionaries, or lists) before they enter the context or checkpoint storage, or implement a custom serialization/deserialization strategy for these object types if they must be persisted. Consider using libraries like `jsonpickle` if you have complex object graphs.
Related AI & LLMs Packages
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
Get up and running with OpenAI gpt-oss, DeepSeek-R1, Gemma 3 and other models.
🦜🔗 The platform for reliable agents.
The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.
LLM inference in C/C++
GPT4All: Run Local LLMs on Any Device. Open-source and available for commercial use.
Subscribe to Updates
Get notified when new versions are released