Security and data handling at Tacari

What Tacari holds

You give us conversations. Customer interviews, calls, notes, documents. We turn them into evidence, themes and problems, and then into documents that cite them.

That means we hold the actual words your customers and prospects said to you. We treat that as the most sensitive data in the product, because for most of our users it is the most sensitive data they own.

Do you train AI models on my data?

No, and neither do the model providers we use.

Tacari uses Anthropic and OpenAI to process your content. Both state that data submitted through their APIs is not used to train their models. Both retain API inputs and outputs for a limited period for abuse monitoring, up to 30 days by default, subject to their own published exceptions.

We publish the specifics, with citations to the providers' own documentation, in our data-handling disclosure. We do not paraphrase their policies, because paraphrasing is how a promise drifts from what the vendor actually said.

We do not train any model on your content. We do not use one customer's data to improve another customer's results.

Can another customer see my data?

No, and we can show you exactly how we know.

Every record in Tacari is scoped to your organisation, enforced in two independent places: row-level security in the database, keyed off your authenticated identity, and explicit filtering in the application. Since August 2026 the database also enforces relationship-level constraints, so a record cannot reference a parent belonging to another tenant even if application code were to try.

That isolation was verified by an independent assessment using real credentials. Two separate organisations, a real signed-in user in one, attempting to reach the other across every path available: 52 database tables, direct API access, join tables, every callable database function, and 25 application endpoints, for both reading and writing. Zero records crossed. The test included a deliberate control case, expected to succeed, which did, confirmation the test setup genuinely worked.

Permissions were tested the same way. Under real administrator and member accounts, 22 separate attempts to escalate privileges were made. All 22 were refused, at three independent layers.

Where does my data live, and who else touches it

Your data is stored in the European Union, in Frankfurt. Encryption in transit is enforced (TLS required, no unencrypted fallback), and your data is encrypted at rest.

The third parties that process your data on our behalf:

If that list changes, we will say so.

How we try not to break things

• Every change runs through automated checks before it can ship, including replaying every database change from scratch on a clean database.
• Content generated by AI is sanitised before it is stored and again before it is displayed. That sanitiser was independently tested against 32 deliberate attack payloads and held against all of them.
• The quality of AI output is checked by an automated evaluation on every change that touches it, so a change cannot quietly make the product worse.
• Every six hours, an automated check confirms that every database migration in our codebase has actually been applied to production. It runs on a dedicated database role that can read only the migration ledger and nothing else.

What we do not have yet

We are an early-stage company, and we would rather be upfront about this than have it surprise you later.

• No SOC 2 or ISO 27001. Not started yet. It's a three to six month process once we do.
• No third-party penetration test. Our security has been independently assessed, but not by a commissioned pen-test vendor.
• No single sign-on or enforced multi-factor authentication. Sign-in is email and password, or a magic link.
• No self-service data export yet. If you want your data out, ask us and we will export it for you manually.
• No uptime guarantee or SLA. We will not offer a number we have not measured.
• No Data Processing Agreement yet. If you need one to buy, tell us and we will prioritise it.

How we handle a problem

If we discover a security issue affecting your data, we'll tell you what happened, what data was involved, and what we're doing about it, even before we have all the answers.

To report something, email [email protected], security researchers included. We'll work with you directly and keep you posted as we fix what you find.