DeepAgents
DeepAgents (Deep Agents) is an open-source agent harness by LangChain: an opinionated agent that runs out of the box and already includes planning, sub-agents, a filesystem, context management and approvals.
- Project
- Deep Agents
- Publisher
- LangChain
- Repository
- github.com/langchain-ai/deepagents
- Documentation
- docs.langchain.com/oss/python/deepagents/overview
- Languages
- Python, TypeScript
- Category
- AI agent frameworks
What is DeepAgents?
Deep Agents is an open-source agent harness – a preconfigured agent with a defined way of working that can be extended, overridden or replaced. Its defaults are tuned for long-horizon, multi-step tasks. The project is built on LangGraph and names Claude Code as its inspiration.
Deep Agents is model-agnostic: it works with any language model that supports tool calling – frontier models as well as open-weight or locally hosted models. Besides Python there is a JavaScript/TypeScript version.
What comes built in
- Sub-agents
- Delegate tasks to agents with their own isolated context.
- Filesystem
- Read, write, edit and search files – locally, in a sandbox or remotely.
- Context management
- Summarise long threads and offload tool outputs.
- Memory
- Pluggable stores for knowledge across sessions.
- Human-in-the-loop
- Approve, edit or reject tool calls before they run.
- Skills and tools
- Load reusable behaviours on demand; connect your own functions or MCP servers.
DeepAgents, LangChain and LangGraph
LangGraph is the graph runtime, LangChain provides a lightweight agent layer on top, and Deep Agents is the more comprehensive harness above that. The layers compose: a custom LangGraph graph can be plugged into a Deep Agent as a sub-agent.
Using DeepAgents effectively in an AI-first business model
The project explicitly follows a “trust the LLM” model: the agent can do anything its tools allow. Boundaries are therefore enforced at the level of tools and sandbox, not by expecting the model to restrain itself.
For an AI-first operating model this means: first decide which tools an agent gets at all and which calls require approval. The built-in approval of tool calls is a suitable mechanism – provided the rules behind it have been decided and documented.
From tool to outcome
In AI-first coaching you clarify which outcomes AI agents should be accountable for in your go-to-market and within which framework. REVAGO then designs, builds and runs your system. You pay for results – not for activity.
Frequently asked questions
What is DeepAgents?
DeepAgents (Deep Agents) is an open-source agent harness by LangChain built on LangGraph. Planning, sub-agents, a filesystem, context management, memory, skills and approval of tool calls come built in. The agent works with any model that supports tool calling.
When should I use DeepAgents instead of LangGraph?
DeepAgents fits when you want to deploy a complete agent with planning, context management and delegation straight away. Using LangGraph directly fits when you want to model every step of a workflow yourself. The two can be combined.
Is DeepAgents safe for business use?
DeepAgents follows a “trust the LLM” model: the agent can do anything its tools allow. Safety comes from restricted tools, a sandbox and approvals for critical calls. These limits have to be defined for the specific use case.
Further reading in the wiki
A neutral overview, not a recommendation of individual providers. REVAGO is not affiliated with the listed projects. Descriptions are based on the projects’ public information; features and licences may change.