Live web search and sources

Real-time web retrieval and verifiable claims checked verbatim against fetched page contents without running container sandboxes.

Grounded web search enables factual verification directly within conversations without requiring an active sandbox container. Citations are verified against exact string matches in fetched source documents.

• Direct web retrieval: Queries search providers (Exa, Tavily, Google, and internal indices) and fetches source pages in real time.
• Byte-accurate quotes: A claim is only marked verified if its supporting quotation matches the verbatim retrieved byte stream.
• Multi-provider routing: Automatically balances query load and falls back across providers when rate quotas are reached.
• Egress filtering: Web crawlers and fetching agents obey configured domain allowlists and privacy boundaries.

Search and scraping operations do not transmit user identities or account tokens to upstream search engines.