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 difficulty of managing long-horizon, multi-step tasks by providing built-in defaults for planning, context management, and task delegation.

How it works

Built on top of LangGraph and LangChain, it acts as an opinionated layer that bundles essential middleware. It allows users to define an agent with a specific model and system prompt, then extends its capabilities through pluggable components. It supports any LLM with tool-calling capabilities and integrates with LangSmith for production-grade tracing and evaluation.

Who it’s for

Developers who want a production-ready agent framework that works out of the box but remains extensible, allowing them to swap models, customize prompts, or replace specific components without forking the codebase.

Highlights

• Sub-agents: Delegate complex tasks to specialized agents with their own isolated context windows.
• Filesystem Access: Read, write, and search across local, sandboxed, or remote backends.
• Context Management: Automatically summarizes long conversation threads and offloads tool outputs to disk to save space.
• Human-in-the-loop: Ability to approve, edit, or reject tool calls before they are executed.
• Model Agnostic: Compatible with frontier APIs, open-weight models, and local LLMs (via Ollama, vLLM, etc.).
• Skills & Tools: Supports reusable behaviors (skills) and custom functions or MCP servers.