Exa uses neural embeddings to understand the meaning behind queries, not just keyword matches. This means your agent finds semantically relevant results even when the customer does not use the exact right terminology. Exa also offers an auto-prompt feature that converts natural language questions into optimized search queries automatically.

You configure this in your Tars gambit settings. Common approaches include triggering Exa search when the agent's confidence is below a threshold, when the customer explicitly asks for current information, or when the question falls outside your knowledge base topics. The agent uses Exa only when configured to do so.

Yes. The contents endpoint accepts a list of URLs and returns parsed text, titles, authors, publication dates, and optional summaries for each page. Your agent can extract the full text from any search result to provide comprehensive answers, not just snippet-level previews.

Yes. Exa's answer endpoint returns a generated response alongside an array of citation objects, each containing the source URL, title, author, and relevant text. Your agent presents the answer with links so customers can verify claims and read further. Full transparency built into every response.

Exa's pricing varies by endpoint. The answer endpoint includes a costDollars breakdown showing charges for search, content retrieval, and individual components like neural versus keyword search. You can control costs by limiting search frequency, using keyword mode for simple queries, and caching common results.

Yes. Exa's search endpoint supports domain filtering and category filtering parameters. You can whitelist only trusted domains, exclude competitor sites, or limit searches to specific content categories. This ensures the agent only returns results from sources you consider authoritative for your use case.

Exa continuously indexes the web, so results include recently published content. You can also use date filtering to restrict results to a specific time window, like the past week or month. This is particularly useful when customers ask about breaking news, recent changes, or time-sensitive topics.

No. Search results and page contents are fetched in real-time during the conversation and used only to formulate the agent's response. Tars does not cache or store the web content returned by Exa. Each query is live, ensuring the customer always gets the most current information available.