Repolode
Submit
langchain-ai

langchain-ai/deepagents

An opinionated, batteries-included agent harness for Python that runs long multi-step tasks out of the box on LangGraph.

Open-source alternative to:Claude CodeCursor

View on GitHubWebsite
Stars
30,441
Forks
4,293
Open issues
211
Last commit
6 hours ago

What is it?

Deep Agents is an agent harness: a preconfigured agent loop that bundles planning, a filesystem, sub-agent delegation, and context management, so developers can start from a working agent instead of wiring these pieces together themselves. Each part can be overridden or replaced without forking the project.

The harness is built on LangGraph and works with any model that supports tool calling, including hosted frontier APIs, open-weight models, and self-hosted models served through Ollama, vLLM, or llama.cpp. Sub-agents run with isolated context windows, and tool outputs can be offloaded to disk to keep long threads manageable. Human-in-the-loop approval, skills loaded on demand, and MCP server support are also included.

A TypeScript version is available as deepagents.js, and a terminal coding agent called Deep Agents Code is built on the same harness.

Key features

  • 01Sub-agents with isolated context windows for delegating tasks
  • 02Pluggable local, sandboxed, or remote filesystem backends
  • 03Context summarization and offloading of tool outputs to disk
  • 04Human-in-the-loop approval, editing, or rejection of tool calls
  • 05On-demand skills and support for any MCP server or custom function as tools

Use cases

  • Building long-running research or report-writing agents that plan and read/write files
  • Creating a self-hosted coding agent on top of a local or open-weight model
  • Adding approval gates to agent actions that touch sensitive tools
  • Composing custom LangGraph graphs as sub-agents inside a managed harness

Quick start

  1. 1

    Install the package with uv

    uv add deepagents
  2. 2

    Create a deep agent in Python and invoke it

Things to know

  • Trusts the LLM model; boundaries must be enforced at the tool or sandbox level.
  • Requires an LLM that supports tool calling, and the quickstart example uses an OpenAI model that needs an API key.

Topics

#agent#llm#langgraph#langchain#python#tool-calling#multi-agent#mcp#ai#deepagents

FAQ

What is deepagents?
An opinionated, batteries-included agent harness for Python that runs long multi-step tasks out of the box on LangGraph. Deep Agents is an agent harness: a preconfigured agent loop that bundles planning, a filesystem, sub-agent delegation, and context management, so developers can start from a working agent instead of wiring these pieces together themselves. Each part can be overridden or replaced without forking the project.
Is deepagents free to use? What license does it have?
deepagents is open source under the MIT license. Check the license terms before using it in commercial products.
Is deepagents still maintained?
Actively maintained. The last commit to deepagents was on Oct 20, 2026.
What can I use deepagents for?
Building long-running research or report-writing agents that plan and read/write files Creating a self-hosted coding agent on top of a local or open-weight model Adding approval gates to agent actions that touch sensitive tools Composing custom LangGraph graphs as sub-agents inside a managed harness
How do I install deepagents?
Install the package with uv: uv add deepagents Create a deep agent in Python and invoke it
Where can I get deepagents?
The source code and installation instructions are on GitHub: https://github.com/langchain-ai/deepagents

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects