🍳 Cookbook: Agentic Search Engine¶

Build an AI-powered search assistant that combines web search, scraping, and LLM summarization into a single intelligent API.

🎯 What You'll Build¶

A service that: 1. Searches the web (Brave, Google, Bing) 2. Scrapes relevant pages for full content 3. Uses AI to summarize findings with citations

🏗️ Phase 1: Clone & Setup¶

1. Get the Code¶

git clone https://github.com/traylinx/agentic-search-engines.git
cd agentic-search-engines

2. Install Dependencies¶

pip install -r requirements.txt

3. Configure Environment¶

cp .env.example .env
nano .env

Key Variables:

# LLM Configuration
LLM_MODEL=openai/gpt-oss-20b
LLM_SUMMARY_MODEL=llama-3.3-70b-versatile
TRAYLINX_BASE_URL=https://switchai.traylinx.com/v1

# Search Providers (get free keys)
BRAVE_API_KEY=your-brave-key
# Optional: GOOGLE_API_KEY, BING_API_KEY

🚀 Phase 2: Run the Service¶

python api.py
# Server running at http://localhost:8000

🧪 Phase 3: Test the Flows¶

Flow 1: Full Agentic Search¶

The complete orchestration: search → scrape → AI summarize.

curl -X POST http://localhost:8000/v1/search \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What are the latest trends in AI agents?",
    "flow": "agentic",
    "params": {
      "use_search": true,
      "providers": ["brave"]
    }
  }'

Response:

{
  "answer": "Based on my research, the latest trends in AI agents include...",
  "sources": [
    {"title": "...", "url": "https://..."},
    {"title": "...", "url": "https://..."}
  ]
}

Flow 2: Search Only¶

Just get raw search results without AI processing.

curl -X POST http://localhost:8000/v1/search \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Python async best practices",
    "flow": "search_only"
  }'

Flow 3: Scrape Only¶

Extract content from a specific URL.

curl -X POST http://localhost:8000/v1/search \
  -H "Content-Type: application/json" \
  -d '{
    "query": "extract main content",
    "flow": "scrap_only",
    "params": {
      "urls": ["https://example.com/article"]
    }
  }'

🤖 Phase 4: OpenAI-Compatible Chat¶

Use it like a drop-in replacement for ChatGPT—with live web access!

curl -X POST http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "agentic-search",
    "messages": [
      {"role": "user", "content": "What is the weather in Tokyo right now?"}
    ]
  }'

The service automatically detects intent: - "Show me videos about X" → Video search - "Extract text from URL" → Web scraping - General questions → Search + Summarize

🔐 Phase 5: Enable A2A Authentication¶

To let other agents call your search service, add Sentinel auth.

1. Get Agent Credentials¶

From console.traylinx.com: - TRAYLINX_CLIENT_ID - TRAYLINX_CLIENT_SECRET

2. Add to .env¶

TRAYLINX_CLIENT_ID=ag-xxx
TRAYLINX_CLIENT_SECRET=ts-xxx
AUTH_SERVICE_URL=https://sentinel.traylinx.com/validate

3. Call as an Agent¶

curl -X POST http://localhost:8000/v1/search \
  -H "X-Agent-Secret-Token: your-token" \
  -H "X-Agent-User-Id: your-agent-id" \
  -H "Content-Type: application/json" \
  -d '{"query": "AI news", "flow": "agentic"}'

🏛️ Architecture Overview¶

          Request
             │
             ▼
     ┌───────────────┐
     │ Intent Router │ ← Detects: search/scrape/video/news
     └───────┬───────┘
             │
     ┌───────┴───────┐
     ▼               ▼
┌─────────┐   ┌───────────┐
│ Search  │   │  Scraper  │
│  Agent  │   │   Agent   │
└────┬────┘   └─────┬─────┘
     │              │
     └──────┬───────┘
            ▼
     ┌─────────────┐
     │ LLM Summary │ ← Generates cited answer
     └─────────────┘

📚 Next Steps¶

Deploy to production with Docker
Add Redis caching for faster repeated queries
Integrate with your own agent via A2A tools