How BizQuery works
BizQuery is built so you can trust what the AI tells you. Three ideas make that possible.
Grounded answers, not guesses
When you ask a question, the advisor doesn’t answer from a model’s general training. It searches your documents for the most relevant passages and answers from those. This is retrieval‑augmented generation (RAG): the model reasons, but over your evidence.
What it searches is set by the project you’re working in — its libraries are the scope for Ask and for the advisors a Decisions playbook runs. If your knowledge has nothing relevant to a question, a good advisor says so rather than inventing an answer.
Citations you can check
Every answer carries its evidence — the passages and documents the advisor pulled in, with the tool calls it made. Expand the collapsible sources strip to confirm a claim came from your source material, and click any citation to open the document at the exact page.
Human sign-off for decisions
A question is one thing; a decision is another. Decisions implement maker‑checker: the AI drafts a recommendation (the maker), and a human approves, refines, or rejects it (the checker) before it counts. A non‑admin can’t sign off their own request — a distinct checker must — while admins may approve their own. The decision, the reasoning, and the reviewer are all recorded for audit. A playbook’s plan routes through the same gate: nothing with an external effect runs until a human approves it.
Choosing the model
Where your deployment offers a choice, a model picker lets you pick the chat model — a fast model for simple lookups, a stronger reasoning model for analysis. An Auto option routes each question to the right one for you.