hybrid-deep-search（混合深度搜索）

Hybrid Deep Search 🚀

Intelligent three-tier search system that automatically routes queries between Brave API and OpenAI Codex for optimal cost and quality.

Architecture

User Query
   ↓
Query Analyzer (router.py)
   ↓
   ├─→ Simple Questions → Brave API (web_search)     Fast, Free
   ├─→ Complex Questions → OpenAI Codex (gpt-5-codex)  Deep Analysis, Paid
   └─→ Manual Mode → User Specified

Quick Start

1. Install Dependencies

pip install openai python-dotenv requests

2. Configure API Keys

Brave API (Already Built-in)

No extra configuration needed. Uses OpenClaw's built-in web_search tool.

OpenAI Codex API (Optional for Deep Search)

# Get API Key from: https://platform.openai.com/api-keys

export OPENAI_API_KEY="sk-your-openai-api-key"
export OPENAI_BASE_URL="https://api.openai.com/v1"  # Optional

Usage

Auto Mode (Recommended)

python3 scripts/deep_search.py "your query"
# System automatically analyzes complexity and chooses:
# - Simple questions → Brave API
# - Complex questions → OpenAI Codex

Manual Mode

# Quick Search (Brave API)
python3 scripts/deep_search.py "what is OpenClaw?" --mode quick

# Deep Search (OpenAI Codex)
python3 scripts/deep_search.py "compare LangChain vs LlamaIndex" --mode codex

Focus Modes

# Academic Search
python3 scripts/deep_search.py "AI agent frameworks research" --mode codex --focus academic

# News Search
python3 scripts/deep_search.py "latest AI news" --mode quick --focus news

# General Web Search
python3 scripts/deep_search.py "OpenClaw documentation" --mode quick --focus web

Parameters

Complexity Routing Rules

Auto mode routes based on query analysis:

→ Brave API (quick)

• Simple factual queries (what/who/when/where)
• Definition lookups
• Quick fact-finding
• Single-topic searches

Examples:

• "what is OpenClaw?"
• "who created Python?"
• "latest AI news today"

→ OpenAI Codex (codex)

• Comparison analysis
• Deep reasoning
• Multi-source synthesis
• Complex questions
• Requires inference/summarization

Examples:

• "compare LangChain vs LlamaIndex in detail"
• "analyze impact of AI on job market"
• "explain quantum computing applications in healthcare"

Cost Optimization

Brave API

• ✅ Completely free
• ⚡ Fast response (<2s)
• 📊 Controllable result count

OpenAI Codex (gpt-5-codex)

• 💰 Usage-based pricing
• 🧠 Deep reasoning capabilities
• ⏱️ Slower response (5-30s)
• 💡 Free tier available for new users

Recommendation: Use auto mode to let the system optimize costs for you.

Technical Details

Query Analyzer (router.py)

Analyzes query complexity using NLP rules:

• Keyword detection (compare/analyze/explain...)
• Sentence length
• Complexity scoring
• Automatic routing decisions

Brave API Integration

Uses OpenClaw's built-in web_search tool:

• Called via Bash tool
• Handles requests automatically
• No extra authentication needed

OpenAI Codex Integration

• Uses gpt-5-codex model
• Built-in web search tool
• OpenAI Chat Completions API format

Example Use Cases

Case 1: Quick Fact Lookup

python3 scripts/deep_search.py "OpenClaw version 2026"
# → Auto-uses Brave API
# → Result: Fast return, free

Case 2: Deep Analysis

python3 scripts/deep_search.py "comprehensive analysis of AI agent architectures"
# → Auto-uses OpenAI Codex
# → Result: Deep analysis, multi-source synthesis

Case 3: Academic Research

python3 scripts/deep_search.py "recent papers on multi-agent systems" --mode codex --focus academic
# → Uses OpenAI Codex
# → Result: Focused on academic literature

Advanced Usage

Batch Search

# Create queries.txt
echo "query 1" >> queries.txt
echo "query 2" >> queries.txt

# Batch execute
for query in $(cat queries.txt); do
  python3 scripts/deep_search.py "$query" --mode auto
done

Format Output

# JSON output
python3 scripts/deep_search.py "query" --format json

# Markdown output (default)
python3 scripts/deep_search.py "query" --format markdown

# Plain text output
python3 scripts/deep_search.py "query" --format text

Troubleshooting

Brave API Not Responding

# Check OpenClaw web_search tool
# No extra configuration needed

OpenAI Codex Authentication Failed

# Check environment variable
echo $OPENAI_API_KEY

# Reset
export OPENAI_API_KEY="sk-..."

Python Dependencies

pip install --upgrade openai python-dotenv requests

Complexity Scoring System

Scoring Factors (0-10 total)

1. Keyword Matching (+6)

- compare/analyze/explain/why/how...

1. Query Length (+2)

- >15 words = +2 - >8 words = +1

1. Question Pattern (+1)

- Complex wh-questions

1. Technical Terms (+1)

- API/framework/architecture...

1. Simple Keyword Penalty (-2)

- what is/who is/list of...

Decision Thresholds

• 0-2 score: Brave API (quick)
• 3+ score: OpenAI Codex (codex)

Configuration

Config File

cp config.json.example config.json
# Edit config.json with your settings

Example config.json:

{
  "search_settings": {
    "default_mode": "auto",
    "default_focus": "web",
    "max_results": 10,
    "router_threshold": 3,
    "verbose": false
  },

  "openai_codex": {
    "enabled": true,
    "api_key": "YOUR_OPENAI_API_KEY_HERE",
    "base_url": "https://api.openai.com/v1",
    "model": "gpt-5-codex",
    "max_tokens": 4096,
    "temperature": 0.7
  }
}

Credits

• Built by Office_bot
• Powered by OpenClaw

References

• Brave Search API
• OpenAI GPT-5-Codex
• OpenAI API Docs