LangChain Community

langchain-community is the package that historically hosted LangChain’s long tail of community-contributed integrations — chat models, LLMs, embeddings, vector stores, document loaders, tools, retrievers, and callbacks for hundreds of third-party providers — kept separate from LangChain’s core abstractions so the core package could stay lean and stable.

As of this listing, the maintainers have announced that langchain-community is sunset: no new integrations or fixes are being accepted, and users are directed toward dedicated per-provider packages (e.g. langchain-openai, langchain-anthropic) or the LangChain integrations ecosystem going forward. The package remains installable and widely used in existing projects, but new projects should consult the current LangChain documentation before adopting it.

What You Get

• Chat model and LLM wrappers for dozens of third-party providers not bundled into LangChain core
• Vector store integrations (langchain_community.vectorstores) for a wide range of embedding databases
• Document loaders, retrievers, and tools for connecting LangChain chains/agents to external data sources and APIs
• Callback handlers and chat-history loaders for tracing and importing conversation data
• A stable interface layer so integrations can evolve independently of LangChain core release cycles

Common Use Cases

• Wiring an existing LangChain chain or agent to a specific third-party chat model, vector store, or tool via a community integration
• Maintaining legacy LangChain applications that were built before integrations were split into per-provider packages
• Loading documents or chat history from a source with an existing langchain_community loader rather than writing a custom one
• Prototyping quickly against a wide range of providers before pinning to a dedicated per-provider package

Under The Hood

Architecture The repository is a single-package monorepo rooted at libs/community, with the importable langchain_community module organized by integration category — chat_models/, embeddings/, vectorstores/, retrievers/, tools/, agents/, callbacks/, storage/, chat_loaders/, cross_encoders/, query_constructors/ — each a directory of per-provider modules implementing LangChain core’s base interfaces (BaseChatModel, VectorStore, BaseRetriever, etc.) for a specific third-party service.

Tech Stack Pure Python, packaged with a pyproject.toml/uv.lock-based build, depending on langchain-core for its base abstractions and, per-integration, on the relevant third-party SDKs (loaded lazily/optionally so the base install stays light).

Code Quality Tests are split into tests/unit_tests and tests/integration_tests, mirroring the module structure so each provider integration has isolated unit coverage plus optional integration tests that require live credentials (declared in extended_testing_deps.txt). Given the package’s sunset status, the maintainers note in the README that no further fixes are being merged, so code quality reflects a frozen, no-longer-actively-hardened snapshot.

API Design Each integration implements the same LangChain core base class for its category (e.g. every vector store subclasses VectorStore), so switching between providers within a category typically requires changing only the constructor/import, not the surrounding chain or agent code — the main ergonomic strength of the community-package split.