Trust / How it works
How it works
We connect to a client's accounting and bank data, reconcile it through software we write, and present it as a secure dashboard plus an AI analyst.
The data moves one way. The AI reads finished numbers and never writes them.
Figures flow from the accounting system, through reconciliation, to the dashboard. The AI analyst reads finished, reconciled numbers and cannot author or change a figure. A write guard at the transport layer enforces this. The analyst makes no automated decisions.
Every figure traces back to its source.
The numbers are inspectable SQL against your own reconciled data. Each figure traces to the source record it came from.
Cross-client isolation is verified by a test.
Each client is a separate database with its own credentials. An automated test asserts that a handler sees only its own client's rows, and that pointing one client's credentials at another's database is rejected. A wrong configuration fails the test rather than reaching production.
Access is managed, passwordless, and denies by default.
Client dashboards use a managed identity provider (Auth0) with passwordless email codes and an allowlist that denies access by default. AI connections authenticate with per-deployment tokens. A startup self-test checks the auth configuration on every deploy.
Least privilege, and secrets kept out of source control.
The AI connects as a read-only database role. Secrets are kept out of source control. Code changes run through an automated test suite in CI before deployment, and deployments are gated and manual.