快速启动

比较有意思的是，这里通过make来管理命令的执行，我觉得特别方便，这是很值得学习的地方，或许我的项目也可以通过make来简化需要执行的命令。

配置

1.克隆DeerFlow仓库

git clone https://github.com/bytedance/deer-flow.git
cd deer-flow

2.运行设置向导

从项目根目录(deer-flow/)运行：

make setup

终端输出

Welcome to DeerFlow Setup!
This wizard will help you configure DeerFlow in a few minutes.

Existing configuration detected.

Do you want to reconfigure? [Y/N]: Y


╔════════════════════════════════════════════╗
║  Step 1/4 · Choose your LLM provider       ║
╚════════════════════════════════════════════╝

   1. OpenAI  (GPT-4o, GPT-4.1, o3)
   2. Anthropic  (Claude Opus 4, Sonnet 4)
   3. DeepSeek  (V3, R1)
   4. Google Gemini  (2.0 Flash, 2.5 Pro)
   5. OpenRouter  (OpenAI-compatible gateway with broad model catalog)
   6. vLLM  (Self-hosted OpenAI-compatible serving)
   7. Codex CLI  (Uses Codex CLI local auth (~/.codex/auth.json))
   8. Claude Code OAuth  (Uses Claude Code local OAuth credentials)
   9. Other OpenAI-compatible  (Custom gateway with base_url and model name)

Enter choice: 3

  → Available models for DeepSeek:
 * 1. deepseek-chat
   2. deepseek-reasoner

Select model [1]: 1


╔════════════════════════════════════════════╗
║  Step 1/4 · Enter your API Key             ║
╚════════════════════════════════════════════╝

DEEPSEEK_API_KEY:
  ✓ Key will be saved to .env as DEEPSEEK_API_KEY

╔════════════════════════════════════════════╗
║  Step 2/4 · Web Search & Fetch (optional)  ║
╚════════════════════════════════════════════╝

 * 1. DuckDuckGo (free, no key needed)  —  No API key required
   2. Tavily  —  Recommended, free tier available
   3. InfoQuest  —  Higher quality vertical search, API key required
   4. Exa  —  Neural + keyword web search, API key required
   5. Firecrawl  —  Search + crawl via Firecrawl API
   6. Skip for now  (agent still works without web search)

Choose a web search provider [1]:

 * 1. Jina AI Reader  —  Good default reader, no API key required
   2. Exa  —  API key required
   3. InfoQuest  —  API key required
   4. Firecrawl  —  Search-grade crawl with markdown output, API key required
   5. Skip for now  (agent can still answer without web fetch)

Choose a web fetch provider [1]:

╔════════════════════════════════════════════╗
║  Step 3/4 · Execution & Safety             ║
╚════════════════════════════════════════════╝

  → Choose how much execution power DeerFlow should have in this workspace.
 * 1. Local sandbox  —  fastest, uses host filesystem paths
   2. Container sandbox  —  more isolated, requires Docker or Apple Container

Execution mode [1]:

  ! Local sandbox is convenient but not a secure shell isolation boundary.
  → Keep host bash disabled unless this is a fully trusted local workflow.
Enable bash command execution? [Y/N]: Y
Enable file write tools (write_file, str_replace)? [Y/N]: Y

╔════════════════════════════════════════════╗
║  Step 4/4 · Writing configuration          ║
╚════════════════════════════════════════════╝

  ✓ Config written to: config.yaml
  ✓ API keys written to: .env

╔════════════════════════════════════════════╗
║  Setup complete!                           ║
╚════════════════════════════════════════════╝

  ✓ LLM:        DeepSeek / deepseek-chat
  ✓ Web search: DuckDuckGo (free, no key needed)
  ✓ Web fetch:  Jina AI Reader
  ✓ Execution:  Local sandbox
  ✓ Bash:       enabled (host bash)
  ✓ File write: enabled

Next steps:
  make install    # Install dependencies (first time only)
  make dev        # Start DeerFlow

Run make doctor to verify your setup at any time.

配置​

配置