LangChain DeepSeek

LangChain DeepSeek is the official integration package connecting DeepSeek’s hosted models to the LangChain ecosystem. It exposes ChatDeepSeek, a chat-model class that speaks DeepSeek’s OpenAI-compatible API while surfacing LangChain features such as tool calling, structured output, streaming, and reasoning-content handling for DeepSeek’s reasoner models.

Built on langchain-core and langchain-openai, it lets you drop DeepSeek into any LangChain or LangGraph application with the same interface used for other providers, so chains, agents, and RAG pipelines can swap in DeepSeek models with minimal code change.

What You Get

• A ChatDeepSeek chat-model class configured via the DEEPSEEK_API_KEY environment variable
• Tool/function calling and structured (Pydantic) output support through the LangChain interface
• Streaming and reasoning-content handling for DeepSeek’s reasoner models
• Drop-in compatibility with LangChain and LangGraph chains, agents, and RAG pipelines

Common Use Cases

• Using DeepSeek chat and reasoning models inside existing LangChain applications
• Building tool-using agents or RAG pipelines backed by DeepSeek
• Swapping DeepSeek in as a provider without rewriting LangChain chain code

Under The Hood

Architecture - The package is a thin partner integration living at libs/partners/deepseek in the LangChain monorepo. Its core is chat_models.py, where ChatDeepSeek subclasses BaseChatOpenAI from langchain-openai and points the OpenAI client at DeepSeek’s API base (api.deepseek.com/v1, plus a beta endpoint). A model-profile registry under data/_profiles supplies per-model capability metadata, and a model_validator wires the DEEPSEEK_API_KEY secret from the environment.

Tech Stack - Pure Python (>=3.10), built with hatchling, depending on langchain-core and langchain-openai (which brings the openai client and pydantic). Distributed on PyPI independently while versioned inside the monorepo; ships a py.typed marker.

Code Quality - By reusing BaseChatOpenAI the package keeps its own surface small and focused, overriding only DeepSeek-specific behavior such as reasoning-content parsing and endpoint selection. It has a tests directory and inherits the monorepo’s CI, linting, and release tooling.

API Design - Excellent ergonomics: install, set DEEPSEEK_API_KEY, and instantiate ChatDeepSeek(model=…) exactly like any other LangChain chat model. Tool calling, structured output, and streaming come through the shared runnable interface, so no DeepSeek-specific boilerplate is required.