The latest release of bytedance/deer-flow is v2.1.0. The GitHub repository's README describes DeerFlow 2.0 as a complete rewrite that shares no code with the 1.x line. Version 2.0 added InfoQuest reader, web, and image search, and split the HTTP inactivity timeout out into a separate setting. However, the provided material does not separately list what changed in v2.1.0.
This project is a super-agent harness that runs research, coding, and content work for anywhere from minutes to hours. It puts a deep agent in charge as the top-level orchestrator, and ships with a sandbox, long-term memory, tools, skills, sub-agents, and a message gateway together. The README also introduces RAGFlow-integrated private knowledge search, browser control, scheduled tasks, and project-scoped document shelves. The Python version required to run it and per-feature performance figures cannot be confirmed from the provided material.
The repository description focuses on a structure and feature set that combine multiple elements to handle long-running work. But the presented material contains no performance benchmarks or adoption case studies. It is therefore difficult to judge actual processing capability or production suitability from the feature list alone. The specific change log for v2.1.0 is also unconfirmed, so what exactly differs in that version cannot be concluded from this material alone.