Security
Your ledger is the most
sensitive file you own.
We treat it that way. Here is precisely how the product handles your data, and what we have not built yet.
Raw rows never reach the model
The analytics engine computes on your data; only aggregates and small samples are ever sent to a language model. A boundary in the code blocks row-level output, and it is covered by tests rather than by policy.
The model cannot write
Every state change the AI proposes becomes a persisted proposal that a person confirms. It cannot alter your data, your thresholds or your close on its own.
Two-factor is mandatory
TOTP multi-factor is required on every login, including Google sign-in. Sessions are revocable — a security stamp in the token invalidates them centrally.
Tenant isolation, adversarially tested
Access resolves through a single chokepoint. A dedicated test suite plays the attacker: probing another tenant returns not-found rather than forbidden, so identifiers cannot be enumerated.
Encryption in transit and at rest
TLS throughout. Credentials for connected sources are encrypted before storage, never logged, and never rendered back to the browser.
A trail that detects tampering
Close events are hash-chained: each entry commits to the one before it, so editing or deleting any row breaks every hash after it. A nightly check verifies the chain and reports where it broke.
Never
What we do not do
with your data.
- We do not train any model on your data.
- We do not sell, share or syndicate it.
- We do not send raw transaction rows to a third-party model.
- We do not keep it after you delete it.
Straight answers
What we do not
have yet.
SOC 2
Not yet certified. We are happy to walk your team through the controls above, and to answer a security questionnaire in detail.
Data residency
Currently hosted in a single region. If you need data to stay in a specific jurisdiction, talk to us before you sign.
Penetration test
No third-party report to share yet. The tenant-isolation suite is internal, and we will show it to you.
Questions from your security team?
Send them over. We would rather answer a hard questionnaire than have you discover the answer later.