Overview
LangChain is a minimal, configurable framework for creating AI agents. Its agent harness surrounds the model loop with prompts, tools, and middleware, allowing developers to compose behavior around the needs of a particular application and its data.
It belongs to a broader open-source stack for designing, testing, and shipping agents, with control over models, tools, context, memory, and orchestration.
Key Features
- Build an agent from a model, custom tools, a system prompt, and middleware.
- Connect to providers including OpenAI, Anthropic, Google, OpenRouter, Fireworks, Baseten, Ollama, Azure, AWS Bedrock, and Hugging Face.
- Extend projects through integrations for model providers, vector stores, retrievers, and other components.
- Trace, debug, and evaluate agents with LangSmith.
How It Fits the Stack
LangChain is designed for developers who want a customizable agent harness. Deep Agents adds a more complete setup for complex, long-running tasks, including planning, subagents, a virtual filesystem, context compression, and long-term memory.
LangGraph provides lower-level orchestration for stateful workflows, with durable execution, streaming, memory, and human-in-the-loop support. Managed Deep Agents supplies managed deployment infrastructure, while Fleet offers a no-code route for building and running agents.
Developer Resources
The documentation includes tutorials, conceptual guides, API references, error codes, release notes, and migration guides. Its programmatic access options cover both product guidance and detailed package references, so compatible AI assistants and code editors can query documentation through MCP servers.