Software for businesses whose logic has to be right.
Tavuni began as a question: if a spreadsheet already runs a business, can you turn it into software and prove the software computes the same answers? The answer became a compiler, a verification engine and the platform they feed.
How Tavuni is built.
The model is the source of truth
Business logic stays explicit, inspectable and versioned. The application is compiled from it, never the other way round.
Evidence over assertion
A claim needs a test, a run or a screen behind it — and a result that cannot be measured says “not measurable”, not 0%.
AI proposes, people decide
Models are useful and fallible. They draft; deterministic code validates; a person publishes.
Limits stated in the open
What is not done is written down beside what is — including on this website.
The honest status, October 2026.
Built and verified by automated tests
- Spreadsheet compiler and two-level verification
- Business Graph: objects, connections, calculated values, numbering
- Generated application screens, lists, reports and exports
- Stages, workflows, approvals including parallel sign-off
- Roles, record and field access, tenant isolation, tenant addresses
- Audit chain, field history, environment promotion and rollback
- Single sign-on (OIDC, SAML), SCIM, multi-factor authentication
- REST API with scoped keys, signed webhooks, OAuth connectors
- AI governance: policy, budget, metering, provenance of every fact
One real process to pilot.
The engineering question is answered; the commercial one is not. The most useful thing you can send us is a real workbook or procedure that runs part of your business — and the question of whether Tavuni can verify it.
See it work on a real workbook, in your browser.
The example operations workbook opens in Tavuni with no account and nothing uploaded. Every number on this site comes from that same pipeline.