Deep Research Agent 🔬

An autonomous AI research agent that performs deep, multi-step web research with sub-agent delegation, full-page content analysis, and structured report generation.

Features

• 🧠 Autonomous planning — breaks research questions into focused sub-tasks
• 🔍 Deep web search — Tavily-powered search with full-page content extraction
• 👥 Sub-agent delegation — parallel research units with isolated context
• 📄 Structured reports — generates cited, professional-quality reports
• ⚡ Configurable — supports multiple LLM backends and search depths

Architecture

User Query
    │
    ▼
┌──────────────────────┐
│  Orchestrator Agent  │   ← Plans, delegates, synthesizes
├──────────────────────┤
│  tavily_search       │   ← Discovers URLs + fetches full content
│  write_todos         │   ← Task planning
│  read/write_file     │   ← Persistent findings
├──────────────────────┤
│  Sub-Agent: Research │   ← Parallel topic-level research
└──────────────────────┘
    │
    ▼
/final_report.md (cited output)

Usage

With This OpenClaw Skill

Just say: "帮我深度研究一下 [主题]" — the agent will automatically use this skill.

Standalone (Python)

# Install dependencies
pip install -r backend/requirements.txt

# Set API keys
export TAVILY_API_KEY="your_key"         # https://tavily.com/
export ANTHROPIC_API_KEY="your_key"      # or GOOGLE_API_KEY

# Run research
python backend/agent.py "Compare Grok 3 vs Claude Sonnet 4 for coding"

CLI Options

python backend/agent.py "Your topic" \
    --model "anthropic:claude-sonnet-4-5-20250929" \
    --output "my_report.md"

LLM Backends

This agent supports multiple models via LangChain:

| Model | Env Var | Identifier | |-------|---------|------------| | Claude Sonnet 4.5 | ANTHROPIC_API_KEY | anthropic:claude-sonnet-4-5-20250929 | | Claude Sonnet 4 | ANTHROPIC_API_KEY | anthropic:claude-sonnet-4-20250514 | | GPT-4o | OPENAI_API_KEY | openai:gpt-4o | | Gemini 2.0 Flash | GOOGLE_API_KEY | google:gemini-2.0-flash | | Qwen (Groq) | GROQ_API_KEY | groq:qwen-2.5 |

Set via --model flag or DRA_MODEL env var.

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | TAVILY_API_KEY | ✅ | Web search (free tier: 1000 queries/mo) | | ANTHROPIC_API_KEY | one of | LLM backend (recommended) | | GOOGLE_API_KEY | one of | LLM backend (alternative) | | DRA_MODEL | ❌ | Override default model | | TAVILY_MAX_RESULTS | ❌ | Results per search (default: 5) |

Tech Stack

• LangChain DeepAgents — Hierarchical multi-agent research framework
• Tavily — AI-optimized web search
• httpx + markdownify — Full-page content extraction

License

MIT