Deep Agents
github.com/langchain-ai/deepagents- Category
- AI Agents
- Rank
- No. 702Tools index
- Pricing
- Open Source
- Platform
- cli
- Type
- TOOL
- Builder
- langchain-ai
- GitHub
- 28.6k stars
- Latest release
- deepagents-talon==0.0.5
- Date
About
Deep Agents is LangChain's open-source agent harness that ships with built-in filesystem access, subagent delegation, context/memory management, and human-in-the-loop controls for building long-horizon, multi-step LLM agents. It sits atop LangGraph and LangChain's create_agent, works with any tool-calling model, and includes a companion terminal coding agent (Deep Agents Code).
What it can do
Provide agents with filesystem access for reading and writing files
File operations requested by agent → File system changes
Delegate tasks to subagents
Task description → Subagent execution results
Manage context and memory across long-horizon agent tasks
Conversation/task history → Persisted context/memory state
Enable human-in-the-loop approval controls during agent execution
Pending agent action → Human approval/rejection decision
Run a terminal-based coding agent
Coding task or command → Code changes or terminal output
Why it made the leaderboard
It gives you a Claude Code-style agent loop as a library rather than a product: sub-agent delegation, filesystem/shell access, context compaction, and approval gates work out of the box with any tool-calling model, local or frontier. Useful when you want to embed a capable long-horizon agent in your own app instead of scripting one on raw LangGraph.
Intel on Deep Agents
Tags
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.