langchain-ai/deepagents

What it solves

Deep Agents provides a "batteries-included" agent harness that eliminates the need to build complex agentic behaviors from scratch. It solves the problem of setting up long-horizon, multi-step workflows by providing an opinionated set of defaults for planning, context management, and tool usage out of the box.

How it works

Built on top of LangGraph and LangChain, Deep Agents acts as a high-level layer that bundles essential agent middleware. It allows users to create agents that can delegate tasks to sub-agents with isolated contexts, interact with pluggable filesystems (local, sandboxed, or remote), and execute shell commands. It is model-agnostic, supporting any LLM with tool-calling capabilities, and includes built-in mechanisms for summarizing long conversation threads and persisting memory across sessions.

Who it’s for

Developers who want a production-ready agent framework that provides immediate functionality (like filesystem access and sub-agent delegation) without having to manually orchestrate every step of the agent's logic using lower-level tools like LangGraph.

Highlights

• Sub-agents: Delegate complex tasks to specialized agents with their own context windows.
• Context Management: Automatically summarizes long threads and offloads tool outputs to disk to maintain efficiency.
• Pluggable Filesystem: Read, write, and search files across various backends.
• Human-in-the-loop: Ability to approve, edit, or reject tool calls before execution.
• Model Agnostic: Compatible with frontier APIs, open-weight models, and local LLMs via providers like Ollama or vLLM.