30.4K ·GitHub
What is this?
Deep Agents is an open source, batteries-included agent harness from LangChain: an opinionated agent that runs out of the box with sub-agents, a filesystem, context management, and skills bundled in — and every piece can be extended, overridden, or replaced without forking.
Deep Agents is an open source, batteries-included agent harness from LangChain: an opinionated agent that runs out of the box with sub-agents, a filesystem, context management, and skills bundled in — and every piece can be extended, overridden, or replaced without forking.Is this a library I build into my own code, or a finished app?A library: Python here, with deepagents.js as the JavaScript/TypeScript version. Under the hood it runs on the LangGraph runtime, which gives it streaming, persistence, and checkpointing for production deployments.
A library: Python here, with deepagents.js as the JavaScript/TypeScript version. Under the hood it runs on the LangGraph runtime, which gives it streaming, persistence, and checkpointing for production deployments.How do I get my first agent running?Follow the README quickstart: a freshly created Deep Agent can already plan, read and write files, and manage its own context. From there you add tools, swap models, customize prompts, and configure sub-agents per the docs.
Follow the README quickstart: a freshly created Deep Agent can already plan, read and write files, and manage its own context. From there you add tools, swap models, customize prompts, and configure sub-agents per the docs.Do I need an OpenAI or Anthropic key, or will any model do?Any model that supports tool calling works — frontier APIs like OpenAI, Anthropic, and Google, hosted open-weight models on providers such as Baseten or Fireworks, or self-hosted ones via Ollama, vLLM, and llama.cpp. Use any LangChain chat model.
Any model that supports tool calling works — frontier APIs like OpenAI, Anthropic, and Google, hosted open-weight models on providers such as Baseten or Fireworks, or self-hosted ones via Ollama, vLLM, and llama.cpp. Use any LangChain chat model.It can run shell commands and edit files — what stops it from doing damage?Nothing by default: Deep Agents follows a 'trust the LLM' model, so the agent can do anything its tools allow. Enforce boundaries at the tool or sandbox level, and use the built-in human-in-the-loop mode to approve tool calls before they run.
Nothing by default: Deep Agents follows a 'trust the LLM' model, so the agent can do anything its tools allow. Enforce boundaries at the tool or sandbox level, and use the built-in human-in-the-loop mode to approve tool calls before they run.So do I need to write code to use it?Yes — it's aimed at developers composing agents in Python or TypeScript with their own tools and models. If you're starting out, the repo has working examples, and the LangChain team runs free LangChain Academy courses.
Yes — it's aimed at developers composing agents in Python or TypeScript with their own tools and models. If you're starting out, the repo has working examples, and the LangChain team runs free LangChain Academy courses.What comes bundled out of the box?Sub-agents with isolated context windows, a pluggable filesystem (local, sandboxed, or remote backends), context management that summarizes long threads and offloads tool outputs to disk, shell access, persistent memory across sessions, human-in-the-loop approvals, loadable skills, and any MCP server as a tool.
Sub-agents with isolated context windows, a pluggable filesystem (local, sandboxed, or remote backends), context management that summarizes long threads and offloads tool outputs to disk, shell access, persistent memory across sessions, human-in-the-loop approvals, loadable skills, and any MCP server as a tool.How is it different from LangChain or LangGraph?They're layers of one stack: LangGraph is the graph runtime, LangChain's create_agent is a minimal harness, and Deep Agents is the opinionated full harness on top with planning, context management, and delegation included. The layers compose — any LangGraph graph can be passed in as a sub-agent to a Deep Agent.
They're layers of one stack: LangGraph is the graph runtime, LangChain's create_agent is a minimal harness, and Deep Agents is the opinionated full harness on top with planning, context management, and delegation included. The layers compose — any LangGraph graph can be passed in as a sub-agent to a Deep Agent.What is Deep Agents Code?A pre-built coding agent for your terminal, similar to Claude Code or Cursor, but powered by any LLM you point it at. It installs with: curl -LsSf | bash.
A pre-built coding agent for your terminal, similar to Claude Code or Cursor, but powered by any LLM you point it at. It installs with: curl -LsSf | bash.Why build this when Claude Code exists?That's the stated inspiration: Deep Agents is an attempt to identify what makes Claude Code general-purpose and push it further — as an open source harness that works with any LLM and lets you override any piece.
That's the stated inspiration: Deep Agents is an attempt to identify what makes Claude Code general-purpose and push it further — as an open source harness that works with any LLM and lets you override any piece.What are similar projects?deer-flow, hive, DeepCode, SystemOneHarness, langgraph, powercontext, open-multi-agent, Orbital
deer-flow, hive, DeepCode, SystemOneHarness, langgraph, powercontext, open-multi-agent, OrbitalWhich topics is it in?deepagents, langgraph, mcp-client, harness-engineering, long-term-memory, human-in-the-loop, agent-harness, langchain, context-engineering, harness, agent-framework, local-llm
deepagents, langgraph, mcp-client, harness-engineering, long-term-memory, human-in-the-loop, agent-harness, langchain, context-engineering, harness, agent-framework, local-llm