DeerFlow: FULLY FREE Local DEEP Research Agent
Source: YouTube — WorldofAI, published 2025-05-31 Tool covered: DeerFlow
Summary
DeerFlow is ByteDance’s open-source deep-research framework — a self-hostable LangGraph-based multi-agent system that mimics the Manus / GenSpark / OpenAI Deep Research workflow with a Coordinator → Planner → Research Team (agents + coder) topology. Notable for being a frontier-lab open-source release (ByteDance is the wiki’s first first-party Chinese-megacap contribution). Optimized for deep reasoning models (DeepSeek R1, Qwen) but works with any OpenAI-compatible endpoint including Ollama, LM Studio, and OpenRouter free tiers.
Architecture
The Coordinator → Planner → Reporter pattern is the load-bearing idea:
- Coordinator receives the user prompt, decides scope
- Planner decomposes into multi-step research plan
- Research Team — agents (search/MCP/RAG) and a coder agent execute each step in parallel
- Reporter stitches outputs into final brief, including citations, images, charts
Supports human-in-the-loop checkpoints between phases. Built-in: 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.
What’s notable
- First wiki entry from ByteDance — Chinese megacap publishing a serious open-source agent framework
- Built on LangGraph — adds another data point to the LangChain org’s downstream impact
- Local-first by default — Ollama/LM Studio support is first-class, not an afterthought
- Targets the same use case as Archon OS but biases toward research-output formats (briefs, podcasts) vs Archon’s code-task-orchestration framing