Deepagents is an open-source agent harness built on LangChain and LangGraph. It equips agents with planning tools, a filesystem backend, sub-agent spawning, and middleware for handling complex, long-horizon tasks reliably and cost-effectively.
Deepagents is a powerful, open-source framework designed to empower AI agents to tackle long-horizon tasks that span dozens of tool calls. Inspired by state-of-the-art systems like Claude Code and Manus, it implements key principles such as planning before execution, computer access via filesystem and shell tools, and sub-agent delegation for isolated task handling. Built on LangChain and LangGraph, it compiles into a flexible StateGraph that supports streaming, human-in-the-loop (HITL), memory, and LangGraph Studio integration.
Every agent comes with a standard toolkit:
| Tool | Description |
|---|---|
write_todos / read_todos | Structured task planning and progress tracking |
ls, read_file, write_file, edit_file, glob, grep, execute | Filesystem operations (sandboxed shell optional) |
task | Spawn sub-agents for delegation |
from deepagents import create_deep_agent
agent = create_deep_agent(system_prompt="Conduct research and write a report.")
result = agent.invoke({"messages": [{"role": "user", "content": "What is LangGraph?"}]})```
## Resources
- [Documentation](https://docs.langchain.com/oss/python/deepagents/overview)
- [Quickstarts](https://github.com/langchain-ai/deepagents-quickstarts)
- [CLI](libs/deepagents-cli/)
With 6.8k+ stars and recent PyPI release (Nov 2025), it's rapidly gaining traction for scalable agentic AI.