The problem
How it works
DeerFlow began as a deep research framework; version 2.0 is a rewrite built on LangGraph and LangChain that the README calls a super agent harness. It gives the lead agent a filesystem, memory, skills and sandbox-aware execution, and lets it plan and spawn sub-agents for multi-step work. It also connects to MCP servers, scheduled tasks and chat channels such as Slack, Telegram and Discord. A setup wizard writes the config, and the app runs locally through Docker or a local toolchain.
Results and limitations
The README describes the design and setup rather than research results. Its only outcome claim is about popularity: it says the 2.0 launch reached the top of GitHub Trending on 28 February 2026. No benchmark or quality figures are given.
Before you try it
The README states that DeerFlow can run commands and read and write files, is meant for a local trusted environment, and that a gateway admin account amounts to code execution on the host, so do not expose it to a network without reading its security notice. It works best with long-context models (100k tokens or more) with reasoning and tool use, which can make long runs costly. The README highlights a ByteDance search toolset and recommends models from its own cloud partner, so compare options before committing. Check cited pages yourself before relying on a report. Results depend on your setup and were not independently reproduced by Top 200 Repos.
Tools and repositories
- Tools
- DeerFlow 2.0 (Docker or local) · An OpenAI-compatible model with long context · A web search and crawling tool
- Prerequisites
- Docker, or Python 3.12 and Node.js 22 with pnpm, uv and nginx, plus model and search API keys.
Try it yourself: Commission a cited research brief
Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].
Research this question: [Research question]. First show me a short plan: the sub-questions you will investigate, the kinds of sources you will use and roughly how many searches you expect, then wait for my go-ahead. Prefer primary sources such as official documentation, papers and filings, and use no more than [Source limit] sources. In the final brief for [Who will read it], attach a link to every factual claim, mark anything you could not confirm as unverified, and list open questions separately. Only read public pages, respect site terms and do not log in to any account, submit forms or download files without asking me.Spot-check the cited links before sharing the brief.
Related videos
About bytedance/deer-flow
DeerFlow 2.0: ByteDance's OpenClaw Rival (opens in a new tab)
The description links the GitHub repository.
Watch on YouTube (opens in a new tab)About bytedance/deer-flow
DeerFlow + Ollama: Super Agent Harness That Can Do Almost Anything (opens in a new tab)
The description links the GitHub repository.
Watch on YouTube (opens in a new tab)Independent creators. Listing a video implies no endorsement, sponsorship or partnership.
Sources
- bytedance/deer-flow README (opens in a new tab)
Project documentation · github.com · Undated · Source checked 12 Nov 2026, 03:30 UTC
What this source supports
- DeerFlow 2.0 is a ground-up rewrite built on LangGraph and LangChain, described as a super agent harness
- Provides a filesystem, memory, skills and sandbox-aware execution, and plans and spawns sub-agents
- Model-agnostic through OpenAI-compatible APIs; works best with 100k+ token context, reasoning and tool use
- Security notice: runs commands and edits files, designed for a local trusted environment; gateway admin equals host code execution
- Claims the 2.0 launch reached the top of GitHub Trending on 28 February 2026; MIT licence