ByteDance dropped DeerFlow 2.0 last week — it hit #1 GitHub Trending within 48 hours and most engineers haven’t looked at it yet.
Unlike n8n or Dify, DeerFlow runs code inside an isolated Docker sandbox per session. Shell, browser, filesystem. The agent has an actual computer to work in.
Setup is three commands. You need Docker, a Tavily API key, and one LLM provider (OpenAI / Anthropic / DeepSeek / Ollama all work).
Budget pick: DeepSeek V3 at $0.0014/1K tokens — 50× cheaper than GPT-4o, negligible quality gap on research tasks.
Three things to know upfront:
<ul>
<li>8–15s cold start per session (container pool management not yet implemented)</li>
<li>Memory is per-session by default — Redis backend needed for cross-session continuity</li>
<li>API key sprawl is real — use a secrets manager from day one</li>
</ul>
Full setup guide with config.yaml examples, model comparison table, and production trade-offs: <a href="https://www.theinfinity.dev/articles/self-hosting-deerflow-2-complete-setup-guide" target="_blank" rel="noopener nofollow">https://www.theinfinity.dev/articles/self-hosting-deerflow-2-complete-setup-guide</a>
#docker #selfhosted #aiagents #opensource #devops #llm #bytedance #langchain

🚀 Daily Open Source Tools Squad delivers daily articles, open-source projects, and technical resources for Developers & DevOps pros. Stay updated with the latest trends & insights!

📬 Join our mailing list! → theinfinity.dev

Daily Open Source Tools

ByteDance released DeerFlow 2.0, an AI agent framework that runs code in isolated Docker sandboxes per session, giving agents access to shell, browser, and filesystem. Setup requires Docker, a Tavily API key, and an LLM provider. Key limitations include 8–15s cold starts, per-session memory by default (Redis needed for persistence), and API key management complexity. DeepSeek V3 is highlighted as a cost-effective model option at $0.0014/1K tokens. The post links to a full setup guide with config examples and production trade-offs.

DeerFlow 2.0 Setup Guide