AI data handling

Why

An AI that reads your enquiries holds the commercially sensitive core of your business.

How

The AI reads and ranks; your own catalogue, formulas and rules price.

One provider, no training

Kabaido's AI runs on Anthropic models only. Anthropic's commercial terms bar training its models on customer content, and Kabaido does not train on your data either. No other AI vendor receives it.

What is sent, and what is not

Catalogue search and matching run inside the platform, on the database. Making your catalogue searchable sends no product data to the AI provider.

Cite or abstain

Every extracted value records where in your document it came from. A value the source does not state stays empty and raises a clarification. Fit scores, match bands, prices and engine calculations are deterministic code, so the same enquiry on the same catalogue gives the same answer tomorrow.

The homepage AI

• The domain box on the Kabaido homepage runs on its own capped key, separate from the one that serves workspaces, with a daily spend cap and a switch to turn it off.
• It reads only a public website: the home page and at most a few linked pages. It touches no organisation and no workspace data.
• It keeps each domain's answer for 30 days, stores a visitor's address only as a salted hash and is rate limited per visitor and overall.

Spend protection

Every AI call is recorded with its model and cost. The month's usage, an optional daily budget and per organisation rate limits stop a runaway document or a looping integration.