# DeerFlow

DeerFlow is an open-source agent harness that orchestrates sub-agents, memory and sandboxes with extensible skills to carry out research, coding and other long-horizon tasks lasting minutes to hours.

Version 2.0 is a ground-up rewrite that shares no code with the original Deep Research-focused v1, which remains maintained on its own branch. The harness coordinates sub-agents, a sandboxed execution environment, and long-term memory, and can be extended with new skills; a sister desktop tool lets a developer prototype agent ideas, inspect each harness step and replay failures.

It integrates a search and crawling toolset for its research work, supports Docker or local development deployment, an MCP server, messaging-app channels, and tracing through LangSmith, Langfuse or Monocle. Claude Code integration lets a session set goals, compact context manually, delegate to sub-agents and run a terminal workbench.

A one-line prompt can hand the setup instructions to a coding agent so it clones and configures the project itself. Python, MIT licensed.

Website: https://deerflow.tech