DeerFlow lets AI handle research tasks end-to-end with local agents
DeerFlow is an open-source project from ByteDance that turns AI into a ‘dedicated computer’ for research and document creation. It runs sub-agents, memory, and sandboxes together, with built-in skills like report writing, slide creation, and image/video generation. The latest version is published on GitHub, with 83,355 stars and an MIT license. It requires an OpenAI-compatible API key and at…
Key points
- DeerFlow runs AI sub-agents, memory, and sandboxes locally for research and document tasks
- Requires 4 vCPUs and 8GB RAM or Docker with 25GB free space, plus an OpenAI-compatible API key
- Security notice warns against remote access; defaults to local-only access and disables command execution by default
The tool is designed for tasks taking minutes to hours, with a focus on continuous workflows rather than simple Q&A. Users must configure AI providers, web search, and safety settings during setup. The project warns of security risks—it grants strong permissions like command execution and file manipulation—so it should only run in a trusted local environment. A configuration example limits simultaneous sub-agents to three to avoid quota limits, a lesson learned from a failed test with four agents.
How to use DeerFlow: Procedures and precautions for giving AI a 'dedicated computer' to handle research and document creation
note.com · 5 October 2026
Loading the full article…
This text was published by note.com and written by オープンソース研究所. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- AWS adds SageMaker AI inference skill for coding agents · 1 src
- OpenAI's GPT-6 Astra cheats by downloading human bot Stardust in StarSkirmish · 2 src
- Claude Code builds and launches rocket in 5.9-hour hackathon · 1 src
- Anthropic’s MCP lets Claude automate tasks across TradingView and other tools · 1 src
- Simon Willison releases llm-anthropic 0.30 with model refresh and token count commands · 1 src
Comments
via GitHub Discussions