← back to the directory
Deep Agents
LangChainmeasuredA LangChain harness you import rather than run: a small middleware core you compose, on a LangGraph runtime for durable execution, streaming and a human in the loop. Model-agnostic, with a JavaScript sibling in its own repository.
spec
- Maker
- LangChain
- Kind
- framework
- Version
- 0.7.7 · checked 2026-08-20
- License
- MIT
- npm
- Not applicable. It is a Python package.
- Repository
- github.com/langchain-ai/deepagents
getting it
pip install deepagents
how it is built
- Language
- Python
- Ships as
- pip package
- What it can enforce
- Declarative filesystem rules
- Out of the box
- Not recorded yet. We have not opened this one up.
- Extends via
- Middleware stack · Skills and MCP · Subagents
- Embeds via
- Import the library
- Runtime
- LangGraph
- Licence
- MIT
capabilities it claims
Taken from the project’s own documentation. Claimed, not verified by us. Verifying is what the playground is for.
- tools
- memory
- sub-agents
- sandbox
- mcp
- skills
- sessions
- hooks
going deeper
This one is measured. Every concept in its docs is laid out as a place, sized by how much source sits behind it, with a real recorded turn you can step through.
open Deep Agents in the playground