Web Search (Tavily)

Web Search (Tavily)

Search the web and get ranked results with snippets, URLs, and an AI-generated answer summary. Powered by Tavily's search API, optimized for LLM consumption.

How It Works — Under the Hood

This skill wraps the Tavily Search API, which is purpose-built for AI agents and LLMs. Unlike traditional search engines that return HTML pages for humans to browse, Tavily returns clean, structured data optimized for programmatic consumption.

Search Pipeline

1. Query processing — your search query is sent to Tavily's search infrastructure. The query is analyzed for intent and expanded internally to improve recall.

2. Web crawling & indexing — Tavily maintains its own web index (separate from Google/Bing). It crawls and indexes pages with a focus on content quality and freshness. Results are ranked by relevance, authority, and recency.

3. Content extraction — for each result, Tavily extracts a clean text snippet (not just the meta description). This means you get actual page content, not just SEO-optimized summaries.

4. AI answer generation — optionally, Tavily generates a synthesized answer by reading the top results and producing a concise summary. This is useful when the agent needs a direct answer rather than a list of links.

5. Structured response — results are returned as JSON with title, URL, content snippet, relevance score, and publication date. Ready for LLM consumption without any HTML parsing.

Search Depth: Basic vs. Advanced

• Basic (default) — fast search, typically 1–2 seconds. Good for factual queries, current events, and simple lookups.
• Advanced — deeper search with more sources crawled. Takes 3–5 seconds but returns higher-quality results for complex or niche queries. Uses more Tavily API quota.

Topic Modes

• General (default) — searches the full web index. Best for most queries.
• News — restricts results to news sources. Best for current events, breaking news, and time-sensitive queries.

Domain Filtering

You can include or exclude specific domains to focus results:

• include_domains: ["arxiv.org", "github.com"] — only return results from these domains
• exclude_domains: ["pinterest.com", "quora.com"] — filter out low-quality sources

Why Tavily Instead of Google/Bing APIs?

• LLM-optimized output — Tavily returns clean text snippets, not HTML. No parsing needed.
• AI answer synthesis — built-in answer generation saves an extra LLM call.
• No rate limit complexity — Claw0x handles the Tavily API key and rate limiting for you.
• Consistent pricing — pay per call through Claw0x instead of managing a separate Tavily subscription.

Prerequisites

Requires a Claw0x API key. Sign up at claw0x.com and create a key in your dashboard. Set it as an environment variable:

export CLAW0X_API_KEY="your-api-key-here"

When to Use

• User says "search for", "look up", "find information about", "what's the latest on"
• Agent needs real-time or current data (prices, news, events, documentation)
• Any query where the answer requires information beyond the model's training data
• Research pipelines that need to gather information from multiple web sources

Input

Output Fields

Example

Input: { "query": "best practices for AI agent error handling" }

Output:

{
  "answer": "Key practices include implementing retry with exponential backoff, logging structured error context, using circuit breakers for external APIs, and providing graceful degradation paths...",
  "results": [
    {
      "title": "Building Reliable AI Agents: Error Handling Patterns",
      "url": "https://example.com/ai-agent-error-handling",
      "content": "When building autonomous agents, error handling is critical...",
      "score": 0.95,
      "published_date": "2026-02-15"
    }
  ],
  "result_count": 5
}

Error Codes

• 400 — Invalid query or parameters
• 429 — Rate limit exceeded (try again later)
• 502 — Upstream search API error (not billed)

Pricing

Freemium model — 50 free calls per day, then $0.01 per call. Failed calls and 5xx errors are never charged.