Deep Agents

About

Start from a working long-horizon agent instead of a blank agent loop

Run uv add deepagents, then create_deep_agent(model=..., system_prompt=...) and invoke it with a task; the agent can plan, read and write files and manage its own context from the start. Add tools, swap models or define sub-agents as you need them. Things to know: it sits on top of LangChain's create_agent and LangGraph, so drop down a layer when you need a custom graph; any compiled LangGraph graph can be passed in as a sub-agent; LangChain also ships Deep Agents Code, a terminal coding agent built on it.

Features

• Sub-agents with isolated context windows
• Pluggable filesystem: local, sandboxed or remote
• Context summarizing and tool output offloading
• Human-in-the-loop approval of tool calls
• Skills, custom tools and any MCP server
• Python and TypeScript, any tool-calling model

Install