deepagents

What it is and what it does

Deep Agents is a production-ready agent harness built on LangGraph that handles multi-step LLM-powered workflows with minimal boilerplate. It sits between LangChain's lightweight `create_agent` and raw LangGraph, bundling filesystem access (read, write, edit, search), context summarization, sub-agent delegation, shell execution, persistent memory, human-in-the-loop approval, and skill loading into a single opinionated interface. The agent works with any LLM that supports tool calling—frontier models (OpenAI, Anthropic, Google), open-weight models on providers like Baseten or Fireworks, or self-hosted via Ollama, vLLM, or llama.cpp.

You provide a model identifier, custom tools, and a system prompt; the agent handles planning, context management, and delegation. It's designed for production deployment with LangSmith integration for tracing, evaluation, and monitoring. The harness is extensible—override or replace any component without forking—and composes with custom LangGraph graphs as sub-agents for orchestration beyond the defaults.

Use it for:

• Build research or analysis agents that read files, search content, and write summaries without managing context windows manually.
• Deploy multi-step task automation (e.g., data processing, report generation) with sub-agent delegation and persistent state across sessions.
• Prototype and ship LLM applications to production faster with built-in tracing, evaluation, and monitoring via LangSmith.
• Create agents that run shell commands or interact with sandboxed environments while maintaining human approval gates on tool execution.
• Extend agent behavior with reusable skills and custom tools without rewriting the agent loop or forking the library.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Deep Agents is an opinionated agent harness that runs LLM-powered agents out of the box, with built-in filesystem access, context management, sub-agent delegation, and skill loading—extensible at any layer without forking.

Yes, if you are building multi-step LLM agents and want production-ready defaults (filesystem, context management, sub-agents, skills) without building from scratch. The active maintenance, zero known vulnerabilities, MIT license, and LangChain ecosystem backing make it low-risk. Install with caution only if your threat model requires strict LLM output validation—the package trusts the model and enforces boundaries at the tool/sandbox level instead.

Install

deepagents on PyPI

pip install deepagents

uv add deepagents

poetry add deepagents

Installing deepagents

Before you install

Low friction install with a pure-Python wheel. Active maintenance (released 2026-08-13, last commit 2026-08-14) backed by LangChain's team. Depends on langchain, langchain-core, langchain-anthropic, langchain-google-genai, langsmith, packaging, and wcmatch—all established libraries.

License in practice

MIT license permits commercial and private use with minimal restrictions. You may use, modify, and distribute the package freely provided you include the license notice.

Quickstart

pip install deepagents

from deepagents import create_deep_agent

agent = create_deep_agent(
    model="openai:gpt-5.5",
    tools=[my_custom_tool],
    system_prompt="You are a research assistant.",
)
result = agent.invoke({"messages": "Research LangGraph and write a summary"})

Requires Python 3.11 or later. You must provide a model identifier (frontier API, open-weight, or local) and configure the corresponding LangChain chat model integration.

Verify before relying

• Whether the package's security model ('trust the LLM') is suitable for your threat model and whether tool/sandbox enforcement is sufficient for your use case.
• Performance characteristics and latency profile under typical multi-step agent workloads.
• Compatibility and integration patterns with custom LangGraph CompiledStateGraphs as sub-agents.

Package facts

Evidence: deepagents-0.7.6-py3-none-any.whl

Keywords: agents, ai, llm, langgraph, langchain, deep-agent, sub-agents, agentic

Tags

More Python Modules packages

Further reading