What is LangGraph Bigtool?
langgraph-bigtool builds a LangGraph agent that starts with one tool, retrieve_tools. When the agent needs a capability, it searches an index of tool descriptions and only the matching tools are bound for the next model call. The index lives in LangGraph's store, in memory or in Postgres, and uses whichever embedding model you configure.
You can replace the default semantic search with your own function, for example one that returns tools by category from a Literal argument, so retrieval does not have to involve embeddings at all. The result is an ordinary LangGraph graph, so streaming, memory and human-in-the-loop work as usual.
It is maintained in the langchain-ai organisation under the MIT licence and depends on langgraph 0.3 or later.
Is LangGraph Bigtool worth it?
Bigtool is less a framework than a well-made pattern, and the pattern matters more now than when it shipped. An agent wired to a few hundred MCP tools pays for every schema in every prompt and picks the wrong one more often; letting it search first, then bind two or three tools, cuts both problems. The quickstart shows the shape with about 50 functions from Python's math module and gpt-4o-mini.
It is small and readable. create_agent takes a model and a dict of tool IDs to tools, the store holds the descriptions, and a custom retrieve_tools function can apply your own ranking, permissions or category filters. That flexibility is the main reason to use it rather than hand-rolling the loop, and the README links the research it draws on, including the Toolshed and Graph RAG-Tool Fusion papers.
The catch is maintenance. Version 0.0.3 on PyPI dates from June 2025, and every commit since has been a Dependabot bump; the latest, on 15 October 2026, merged a langgraph-sdk update. It still has an audience, with 559 stars and about 17,200 downloads in the last month, but it predates LangChain's 1.0 create_agent and middleware, and the library says nothing about newer approaches such as provider-side tool search or meta-tool gateways like Composio.
Retrieval quality is also on you. Tool descriptions written for humans often embed poorly, and a missed retrieval means the agent believes the tool does not exist. Expect to rewrite descriptions and test recall before trusting it with a large catalogue.
Where it earns its place
An internal-tools engineer at a 300-person company
Give one assistant access to every internal API without putting every tool schema in every prompt
Register each endpoint as a tool, index the descriptions in a PostgresStore, and let retrieve_tools return the top few per request; the cost is an embedding call per search plus fewer prompt tokens per turn.
A support-automation lead
Keep billing and service tools apart so the agent cannot mix them
Replace semantic search with a retrieve_tools function that takes a Literal category and returns fixed tool IDs; no embeddings needed, and the model's choice is constrained by type hints.
Before you commit
- Setup
- pip install langgraph-bigtool plus a chat model and embeddings package; build a tool registry dict, put descriptions in a store, compile the graph. The quickstart runs in minutes with an OpenAI key.
- Integrations
- Any LangChain chat model and embedding provider; LangGraph InMemoryStore or PostgresStore; LangChain tools, including tools converted from MCP servers.
- Data and privacy
- MIT licence; tool descriptions and embeddings live in your own store; nothing is sent anywhere except to the model and embedding providers you choose.
- Limits
- Requires langgraph 0.3 or later; no feature releases since June 2025; retrieval misses fail silently, so the agent may answer without the tool it needed.
What’s good
- Keeps hundreds of tool schemas out of the prompt
- Pluggable retrieval: semantic search, categories or your own logic
- Ordinary LangGraph graph, so persistence and streaming still work
- Postgres store for a shared, persistent tool index
What’s not
- No release since 0.0.3 in June 2025; commits are dependency bumps
- Recall depends on tool descriptions you will probably have to rewrite
- Predates LangChain 1.0's create_agent and middleware
Best for
Python teams on LangGraph whose agents have outgrown the number of tools a model can sensibly see at once.
LangGraph Bigtool alternatives
LangGraph Bigtool: common questions
- Is LangGraph Bigtool worth it?
- A clean reference for tool retrieval that does one job well, from a LangChain repo that has had no feature work since its third release in June 2025. We rate it 3.6 out of 5.
- How much does LangGraph Bigtool cost?
- Free and open source under the MIT licence; you pay for the embedding and chat model calls.
- What are the downsides of LangGraph Bigtool?
- No release since 0.0.3 in June 2025; commits are dependency bumps; recall depends on tool descriptions you will probably have to rewrite; predates LangChain 1.0's create_agent and middleware.
- Who is LangGraph Bigtool best for?
- Python teams on LangGraph whose agents have outgrown the number of tools a model can sensibly see at once.
- Is LangGraph Bigtool safe for company data?
- MIT licence; tool descriptions and embeddings live in your own store; nothing is sent anywhere except to the model and embedding providers you choose.
- What are the best LangGraph Bigtool alternatives?
- The closest alternatives we have reviewed are LangGraph, LangChain and OpenAI Agents SDK.
Answers checked Nov 3, 2026.