LangGraph
github.com/langchain-ai/langgraph- Category
- AI Agents
- Rank
- No. 337Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- langchain-ai
- GitHub
- 39.9k stars
- Latest release
- 1.2.11
- Added
- Jul 4, 2026
About
LangGraph is a low-level framework for building stateful, controllable agents and durable multi-step LLM workflows as graphs, with checkpointing, human-in-the-loop, and persistence.
What it does
LangGraph lets developers describe work as connected nodes that react to channel updates, read shared state, and write results onward. Those definitions compile into an execution engine with synchronous and asynchronous streaming, retries, caching, timeouts, checkpoints, interrupts, and resumable runs.
Why it's ranked here
The case for LangGraph rests on unusually concrete execution controls. Its core handles checkpointed state, failure recovery, nested graphs, streaming, retries, caching, and human intervention. The repository also includes benchmarks, a Python SDK, and deployment tooling. That breadth makes it compelling when agent behavior must survive beyond one request.
What's good
Developers can choose graph composition or a function-oriented workflow style. Tasks support parallel futures, retry and cache policies, while asynchronous tasks also support cancellable timeouts and progress heartbeats. Checkpointers preserve state across runs. The engine exposes both synchronous and asynchronous execution, and the repository checks parity between corresponding SDK clients.
Tradeoffs
This is deliberately low-level infrastructure, so quick agent construction belongs in the higher-level Deep Agents package. Safe in-process cancellation applies only to asynchronous tasks. Checkpointed inputs and outputs must be serializable. Local API development uses Docker Compose with Postgres and Redis, while the supplied server workflow requires LangSmith access or a production license key.
How to use it well
Use LangGraph when a Python team needs explicit control over long-running agent state, branching, subgraphs, retries, streaming, and recovery. Start with its graph or function interface, then add checkpoints where continuity matters. Pair it with LangSmith for observability, evaluation, visual debugging, and managed deployment. Choose Deep Agents when rapid, higher-level agent assembly matters more.
Technical notes+
The runtime in libs/langgraph/langgraph/pregel/main.py implements Pregel-style node execution over channels, with sync and async loops, checkpoint savers, stores, caches, retries, streaming transformers, interrupts, and per-node timeout policies. libs/langgraph/langgraph/func/__init__.py layers the task and entrypoint decorators over that engine, returning futures for tasks and supporting checkpointed prior values. libs/cli/langgraph_cli/cli.py provides the Click CLI for launching a configured API server through Docker Compose. .github/scripts/run_langgraph_cli_test.py exercises that stack with Postgres and Redis and probes its health endpoint. .github/scripts/check_sdk_methods.py compares sync and async SDK class methods through AST inspection. libs/langgraph/bench/__main__.py benchmarks sequential, fanout, subgraph, checkpointed, wide-state, and agent workloads.
Observed
- License
- MIT, identified by the repository README license badge.
- Primary language
- Python.
- Installation
- Published as the langgraph Python package and installed with pip.
- Interfaces
- Python library, Python SDK, Click command-line interface, and configurable API server.
- Python support
- The API server configuration accepts Python 3.11, 3.12, or 3.13.
- JavaScript surface
- The README directs JavaScript and TypeScript users to a separate LangGraph.js repository.
- Repository structure
- A multi-package libs layout contains the core runtime, CLI, and Python SDK.
Read from README.md, Makefile, docs/generate_redirects.py, .github/scripts/check_sdk_methods.py, .github/scripts/run_langgraph_cli_test.py, libs/cli/langgraph_cli/cli.py, libs/langgraph/bench/__main__.py, libs/cli/langgraph_cli/__init__.py, libs/cli/langgraph_cli/__main__.py, libs/sdk-py/langgraph_sdk/__init__.py, libs/langgraph/langgraph/pregel/main.py, libs/langgraph/langgraph/func/__init__.py, libs/sdk-py/integration/graph/__init__.py, libs/langgraph/langgraph/graph/__init__.py.
Intel on LangGraph
Tags
Tech Stack
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.