Verified project record

langchain-ai/deepagents

deepagents is an opinionated agent harness that bundles sub-agents, filesystem, context management, and skills so agents for long-horizon, multi-step work run without assembling those pieces yourself. Any bundled piece can be extended, overridden, or replaced.

Project overview

It packages the machinery that minimal agent harnesses leave you to build—planning support, sub-agent delegation with isolated context windows, filesystem-based context management, and skills—while staying model-agnostic across any tool-calling LLM, including self-hosted models via Ollama, vLLM, or llama.cpp.

Project type: MCP · AI Agent
Use cases: Coding & Development · Automation
Deployment: Refer to project documentation
License: MIT

Best for

AI engineers and developers who need an opinionated agent harness for long-horizon, multi-step work with sub-agents, filesystem, and context management bundled from the start.
Teams that want to bring their own tools, MCP servers, and choice of LLM, including self-hosted models, while keeping the ability to extend or replace any bundled piece.

Key capabilities

Delegate tasks to agents with isolated context windows.
Read, write, edit, or search files over pluggable local, sandboxed, or remote backends; summarize long threads and offload tool outputs to disk.
Bring your own functions or any MCP server as tools; works with any LLM supporting tool calling including self-hosted models via Ollama, vLLM, or llama.cpp; approve, edit, or reject tool calls before they run.

Limitations and risks

deepagents follows a 'trust the LLM' model; enforce boundaries at the tool/sandbox level rather than expecting the model to self-police. Review how tools and sandbox backends are configured before adopting.

Getting started

Add the package with a single uv add install and run a short create_deep_agent snippet documented in the Quickstart; coding is required. You will also need access to an LLM that supports tool calling.

Evidence and sources

GitHub project description: The batteries-included agent harness.
README: **Sub-agents** — delegate tasks to agents with isolated context windows
README: **Model-agnostic** — works with any LLM that supports tool calling: frontier, open-weight, or local
README: uv add deepagents
README: Yes! Deep Agents is built on LangGraph, designed for production agent deployments.

Source updated at: 2026-10-23T18:08:33Z

Open original GitHub project