Getting started

You can be asking grounded questions of your own documents in a few minutes.

1. Create a datasource

A datasource is a named collection of documents — a knowledge base the AI can search.

Open Knowledge, choose + Create datasource on the Overview, give it a name (e.g. “Company policies”), and pick who can see it. As an individual account your datasources are always private to you; in an organization an admin can also share a datasource with the whole org or a department. See Access & visibility for the details.

2. Import a PDF

Open your datasource and drag a PDF onto it (or use the upload button). BizQuery then:

• extracts the text (with OCR fallback for scanned pages),
• splits it into passages and builds embeddings for search,
• gives it a clean, standardized title and assigns tags automatically.

You can watch progress under Knowledge → My imports, and adjust a document’s title or tags at any time.

Tip

Only PDFs are supported today. A typical text PDF is ready to query within seconds; large or scanned documents take a little longer while OCR runs.

3. Ask a grounded question

Open Chat, pick your datasource (or one of its tags) as the scope, and ask. The advisor searches your documents, answers in plain language, and shows the evidence — the exact passages it used — so you can verify every claim.

That’s the core loop. From here:

• Turn a datasource into a reusable Advisor with a fixed role and tone.
• Use Proposals when an answer is a decision that needs human sign‑off.
• Drive all of this from the API.