Sign inStart a workspace

Your customers’ words, in your own database.

It reads support conversations for a living, so how they are handled is the product, not a policy page.

One database per workspace

Your conversations are not rows in a shared table beside another company's. A separate Postgres project, with separate credentials.

Redacted before the first write

Personal detail is stripped on arrival, not after storage. People are counted by a hash, so we can say nine people without holding nine identities.

Every action on the record

Append-only and hash-chained, including who approved what and the exact text that was sent. You can read it; nobody can quietly edit it.

Retention you set

A scheduled hard delete runs on your setting, across conversations and everything derived from them. Erasure for one person on request.

What we will not do

  • Receiving signals from a system never authorizes acting on it — those are two separate permissions, and you grant them separately.
  • No message reaches a customer without you approving the exact words.
  • No training of shared models on your conversations.
  • No per-agent scores, so there is nothing to leak about your team.
  • No silent overwrite of your help center — an approval binds to one revision.

Who else touches the data

DatabaseNeonOne project per workspace, US or EU. Redacted conversation text, findings, the audit trail.
ApplicationVercelHosting. Request data in transit; no customer data at rest.
Model inferenceAnthropicGrading and grouping, on redacted text only.
Sign-inClerkNames and email addresses of the people who use the app.
EmailResend or PostmarkOnly when a workspace configures an address: the weekly review and approval notices.
DeliveryYour helpdesk and SlackNot subprocessors — your own systems, acting under permissions you granted.

Key management and error tracking providers are being added; each is listed before it receives anything, and changes are announced before they take effect.

Ask us a security questionsecurity@jarmer.ai