DeerFlow

ByteDance’s open-source deep-research framework. Self-hostable LangGraph-based multi-agent system with a Coordinator → Planner → Research Team → Reporter topology. Optimized for deep reasoning models (DeepSeek R1, Qwen) but accepts any OpenAI-compatible endpoint, including Ollama, LM Studio, and OpenRouter free tiers.

  • Vendor: ByteDance (the wiki’s first first-party Chinese megacap contribution)
  • Repo: github.com/bytedance/deer-flow
  • Site: deerflow.tech
  • Built on: LangGraph

Architecture

Coordinator (scope) → Planner (multi-step plan) → Research Team (search/MCP/RAG/coder agents in parallel) → Reporter (citations + images + charts)

Human-in-the-loop checkpoints between phases. Built-in capabilities: web search, code execution, RAG, MCP integrations, podcast generation, presentation generation.

Demo

“Brief me on GitHub trending repos” → top-10 repos with overview, analysis, images, and citations in ~2 minutes.

Why it matters

  • First ByteDance entry in the wiki — signals Chinese megacaps are now publishing serious open-source agent frameworks, not just models
  • Adds another data point to LangChain’s downstream impact (Open-SWE is the other LangGraph-built frontier project here)
  • Local-first by default — Ollama / LM Studio integration is first-class, not an afterthought
  • Targets the same use case as Archon OS but biases toward research-output formats (briefs, podcasts, slides) vs Archon’s code-task-orchestration framing

Sources

See Also