Questions from technical teams

Through a dedicated, read-only database credential — enforced at the database layer, not just by the AI.

• The agent's database role is granted SELECT only, scoped to your analytics tables. It cannot insert, update, delete, or alter anything, and the database itself refuses any such statement regardless of what the AI generates.
• Every query runs under a hard timeout, so no runaway query can degrade your database.
• Before execution, generated SQL passes a read-only validation check — anything that is not a single SELECT statement is rejected.
• In a self-hosted deployment, you create and control the credential yourself and can revoke it at any time. We never hold write access to your data.

We assume it sometimes will — and we design for that instead of asking you to trust a number.

• Pattern-routed generation. Questions aren't answered by freeform AI improvisation. Each question is routed to one of a defined set of tested SQL patterns covering ranking, growth, comparison, and trend analysis.
• Automated regression testing. Every release is tested against a growing suite of real business questions (currently 74 distinct business questions across 11 analysis patterns, including multi-turn conversations) and must pass fully before deployment.
• Business-rule guards. Where naive SQL gives a technically-correct but misleading answer, we compute the metric with explicit safeguards. Example: return-rate rankings use the actual rate with a minimum-order threshold, so a product with one order and one return can never top the list.
• Show-the-work transparency. The SQL behind every answer is available to you, so results can always be verified against the query that produced them.

No AI analyst is infallible. Ours shows its work.

We are currently in pilot phase: early customers receive an evaluation deployment, and pilot feedback directly shapes final pricing. The structure will follow the two deployment tiers below.

	Managed cloud	Self-hosted
Where it runs	Our infrastructure (your data synced on a schedule to a read-only analytics copy — your live system is never touched)	Your server or cloud environment
Pricing shape	One-time onboarding + monthly subscription	One-time deployment + ongoing support/updates
Your infrastructure cost	None	Your own hosting (we advise on sizing)
Best for	Fast start, minimal IT involvement	Strict data-residency or compliance requirements

Your data never trains anyone's AI model, at any tier. Beyond that baseline, you choose how strict to go.

• Managed cloud tier. Your production system is never exposed to the internet and we never touch your live database. A small sync agent extracts your data on a schedule (e.g. nightly) into a read-only analytics copy — hosted either by us or, if you prefer, in a cloud database account that you own and we can only read. AI inference is routed exclusively through zero-data-retention (ZDR) endpoints: providers that do not store your prompts or results for any period of time.
• Self-hosted tier. The entire system deploys into your own server or cloud environment, connected to your own database. For customers with strict residency requirements (e.g. Malaysian PDPA), inference can additionally be pinned to in-country infrastructure — or run fully inside your perimeter with a locally hosted model, so that no data of any kind leaves your environment.

We'll walk through the specific data flows for your chosen tier during evaluation, including exactly which fields leave your network (if any) and under what policy.