deepagents

Deep Agents is an open-source agent harness designed to provide a "batteries-included" experience for building AI agents. It is model-agnostic, supporting any LLM with tool-calling capabilities, including frontier APIs and local models via Ollama or vLLM. Key capabilities include: - Sub-agents: Ability to delegate tasks to agents with isolated context windows. - Filesystem Integration: Support for reading, writing, editing, and searching across local, sandboxed, or remote backends. - Context Management: Tools to summarize long threads and offload tool outputs to disk to manage token limits. - Shell Access: Capability to execute commands within a chosen sandbox. - Persistent Memory: Pluggable state and store backends for recall across different sessions. - Human-in-the-loop: Mechanisms to approve, edit, or reject tool calls before execution. - Skills and Tools: Support for reusable behaviors (skills) and custom functions or MCP servers. Built on LangGraph, it supports streaming, persistence, and checkpointing, and integrates with LangSmith for tracing and evaluation.