The agent is configured to recognize when a question falls outside its training scope or requires current information. You can set rules for when Tavily search triggers, such as questions about competitors, pricing, recent events, or any topic where freshness matters. The agent decides contextually based on your configuration.

Basic search returns standard web results quickly using 1 API credit. Advanced search performs deeper crawling, visiting more pages and extracting richer content for 2 credits. Use basic for simple lookups and advanced when the question requires thorough, multi-source research.

Yes. Tavily supports include_domains to limit results to specific sites (like your documentation or trusted sources) and exclude_domains to block unreliable or irrelevant sites. Your agent only surfaces information from sources you trust.

No. Every search is performed live via Tavily's API during the conversation. This ensures customers always receive the most current information available. Results are not stored or reused across different chat sessions.

When enabled, Tavily synthesizes a direct answer from the top search results, similar to a featured snippet. The agent receives both the synthesized answer and the supporting search results, so it can provide a concise response while citing sources if needed.

The agent handles empty results gracefully. It may rephrase the search, broaden the query terms, or let the customer know that no relevant information was found online. It can then offer to connect them with a human agent for further assistance.

Yes. Tavily's include_raw_content parameter retrieves the unprocessed text from visited pages. The agent can extract specific sections, summarize lengthy articles, or find precise answers buried deep within a page, far beyond what a simple snippet provides.

Raw internet access returns noisy, unstructured HTML. Tavily is built specifically for AI agents. It returns clean, structured data optimized for language models, with relevance ranking, optional direct answers, and domain filtering. Your agent gets useful information, not web clutter.