Deep Agents
An open-source Python agent harness for multi-step AI work, built on LangGraph.
- Stars
- 30.9k
- Forks
- 4.3k
- Licence
- MIT
- Language
- Python
- Last update
- 4 days ago
- Type
- Library
What it is
Deep Agents is a Python (and JavaScript/TypeScript) library that gives you a ready-made AI agent with planning, file access, sub-agents, and context management already included. You can swap in any tool-calling language model and override or replace parts of it. It is built on LangGraph, a framework for agent workflows.
- Create an agent that researches a topic and writes a summary to a file
- Delegate subtasks to sub-agents that each have their own context window
- Require human approval before the agent runs selected tool calls
- Give the agent skills it can load on demand or connect tools via MCP servers
- Run shell commands or file operations inside a sandbox you choose
Great for
- Developers building AI agents that run multi-step tasks such as research and writing
- Teams that want a working agent harness they can customize instead of building one from scratch
- People who want to use either hosted or self-hosted language models with the same code
- Developers already using LangChain or LangGraph who want a more opinionated setup
Not ideal for
- Non-programmers looking for a finished app: the README centers on writing Python code, and the terminal coding agent is a separate feature
- Anyone who needs strict guarantees the agent will not misuse tools, since the README says it follows a "trust the LLM" model and boundaries must be enforced at the tool or sandbox level
- Projects that only need a simple single-call LLM request, since the harness adds more structure than that requires
How to get it
- 1
Install with a package manager
You'll need: Python (version not stated in the README) · uv (package manager, used in the README install command) · An API key or local model for the LLM you choose (the README does not specify provider setup details)
Add deepagents to a uv project
uv add deepagents - 2
Run it
- Install the package with uv add deepagents
- Import create_deep_agent from deepagents
- Create an agent with create_deep_agent, passing a model and a system prompt
- Call agent.invoke with a message to run the agent
- 3
Get the most out of it
- Add your own tools as Python functions or connect any MCP server
- Swap the model by changing the model string, using any LangChain chat model
- Configure sub-agents, including LangGraph CompiledStateGraph objects
- Pair with LangSmith for tracing, evaluation, and monitoring in production
- Use the Deep Agents Code terminal coding agent if you want a ready-made interface
Downloads
Strengths
- Ships with planning, filesystem, sub-agent, and context management features out of the box
- Works with any LLM that supports tool calling, including local and open-weight models
- Built on LangGraph, which the README describes as providing streaming, persistence, and checkpointing
Trade-offs
- The README is truncated and does not document every option, so details of configuration depend on the external docs
- Its security model places responsibility on the developer to restrict what tools and sandboxes can do
- It is tied to the LangChain ecosystem, so adopting it means adopting those dependencies
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| deepagents (this) | 30.9k | Library | intermediate | 4 days ago | Developers building AI agents that run multi-step tasks such as research and writing |
| langgraph | 44.3k | Library | intermediate | 4 days ago | Python developers building multi-step AI agents that must survive failures or long waits |
| langchain | 149k | Framework | intermediate | 4 days ago | Python developers building LLM-powered applications |
| crewAI | 60.4k | Library | intermediate | 4 days ago | Python developers prototyping multi-agent workflows |
| pydantic-ai | 21k | Library | intermediate | 4 days ago | Python developers building LLM-powered features that need structured, typed output |
| deer-flow | 84.9k | AI model / agent | advanced | 4 days ago | Developers who want to experiment with multi-agent workflows on their own infrastructure |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.