deepagents
View on GitHubThe batteries-included agent harness.
Deep Agents is LangChain's batteries-included agent harness built on LangGraph: planning, sub-agents, filesystem, context management, memory and skills out of the box. Model-agnostic and extensible, it also ships a prebuilt terminal coding agent (dcode).
Use Cases
Long-horizon multi-step research agentsTerminal coding agent powered by any LLMDelegating subtasks to sub-agents with isolated contextReading, writing, editing and searching project filesContext summarization and offloading tool outputs to diskRunning shell commands inside a sandboxCross-session persistent agent memoryHuman approval/editing of tool calls before executionLoading reusable agent skills on demandConnecting external tools and MCP serversProduction agent deployment with tracing and evaluation
Built With
- Language
- Python
- Frameworks
- LangChain · LangGraph · LangSmith
Tags
agent-harness · multi-agent · sub-agents · context-management · filesystem · persistent-memory · human-in-the-loop · skills · mcp · tool-calling · sandbox · langgraph · coding-agent · streaming · checkpointing