Trustworthy by design, not by promise
AI does the reading. Deterministic systems do the checking. You do the approving. Every posting leaves a record.
Confidence + validators
The LLM never has the final say on numbers. Deterministic validators check arithmetic, dates, and balance.
Review-first & audit trail
Web and Gmail receipts post only after you approve them; a clean phone scan files itself and stays editable. Posts, approvals, and corrections land in an append-only audit log with the actor, time, and source.
Least-privilege Gmail
Opt-in and read-only. We look back a fixed window for receipts, tell you what came in, and you can disconnect the mailbox whenever you like.
Encryption & secrets
TLS everywhere in transit; encrypted at rest by Google-managed keys. Config secrets live in Secret Manager, and our services authenticate by workload identity — no key files anywhere.
Role-based access
Owner, co-admin, bookkeeper, add-receipts, and read-only roles, on top of strict per-org isolation enforced on every read and write.
Your data isn't training data
Your books, receipts, and emails are never used to train models — not ours, not anyone's, not across customers.
We never hold your books hostage
If a trial ends or a subscription lapses, WispBooks goes read-only — it never locks you out. Every view, report, and export keeps working; only new posts and edits pause until you come back.
- The accountant export stays available — a year of expenses as CSV, with the receipt images
- The receipt image behind every expense is kept as evidence
- Disconnect Gmail whenever you like — the token is revoked at Google
WispBooks suggests categories and deduction treatment — it isn't tax advice. Suggested classifications carry confidence scores and evidence, and deduction totals are estimates; confirm tax treatment with your accountant before you file.