Security and trust

Guardrails in the architecture, not in a settings page

Receivables data is sensitive and AI must be accountable. These are the controls ResolveAR AI enforces in code.

Tenant isolation
Every query is scoped to your workspace from the signed-in session. The AI service reads data under row-level security.
Approval controls
Policies decide what needs a person. Approvals are tied to the exact facts they were granted for and expire.
Full auditability
Business actions, including those an agent triggered, are recorded with who, what and a correlation id.
Secure integrations
Signed, replay-protected webhooks, encrypted integration tokens and private file storage with short-lived links.

Controls in detail

The workspace is taken from the authenticated session and active membership, never from a URL, header or request body.
Every workspace-owned query filters by workspace inside the database query.
The AI service connects through a restricted database role under row-level security and cannot read business tables.

HTTP-only session cookies; tokens are never exposed to browser scripts.
Role-based access (Owner, Admin, AR Manager, AR Specialist, Viewer) on every change.
Rate-limited sign-in and AI analysis endpoints.
A separate, separately authenticated console for ResolveAR operators.

Agents have no write tools. They recommend; the system and your team act.
Days overdue, aging, balances and adjustment amounts are calculated deterministically, not by the model.
AI output is validated against the records it was given. Unknown ids or amounts are rejected.
Without configured AI credentials the product says so; it never shows made-up analysis.

Uploads are checked by their real file signature and size, and can be malware-scanned before processing.
Files live in a private bucket and are opened only through short-lived links issued after authorization.
Document text is treated as untrusted. Embedded instructions are flagged, never followed.
Document content is sent to a model provider only when your processing policy explicitly allows it.

E-mail goes only to the address on file, checked against bounces and complaints right before sending.
Customer portal links are stored only as hashes, expire, and can be revoked.
Stripe and e-mail webhooks are signature-verified and processed exactly once.
Accounting data imported from QuickBooks stays read-only in ResolveAR.

Need more detail for a security review? Request a demo and tell us what your review covers; we will walk your team through the architecture.

Ready to get started?

See ResolveAR AI on your own collections workflow in a guided walkthrough.