Skip to main content

The Record and retention

In plain words

Every governed run produces one permanent Record: what was asked, what was checked, what was decided and why, and what it cost. Nobody can edit a Record after it is written, not even us. If a Record cannot be saved, Clara refuses to finish the run rather than work without one.

The two fast levels are not governed runs and do not produce a Record. A compliance check returns a decision_ref for correlation, and a quick answer returns a billable receipt. Both are traceable; neither is an audit artifact. See the four offerings.

This page is for everyone, with schema links for developers.

What is inside a Record

Records also note which surface the conversation came from (for example the web widget) and, when a persona was applied, which version of it.

Why "cannot be edited" matters

A record you can quietly change is not evidence. Clara's records are written once, to storage that does not allow overwrites. When you show a record to an auditor or a customer, you are showing what actually happened, not a tidied-up version of it.

Finding your Records

  • In the portal: the Activity page lists every conversation. Open one and the record appears in the side panel.
  • From the widget: each finished conversation shows its record id. Look it up in the portal.
  • Over the API: GET /v1/records lists your organisation's records, and GET /v1/records/{run_id} fetches one.

How long records are kept

During the beta, records are kept under a default retention policy. When a record is deleted at the end of its life, the deletion itself produces a deletion certificate: a signed statement of what was removed, when, and under which policy. So even deletion leaves proof. Per-organisation retention windows arrive with general availability.

Records and model development

Your records are not the same thing as the data behind Clara's model development. Records exist so you can prove what happened; the separate, generalized export used for model development is opt-out and default on. See Model development and your data for what is shared, what is not, and how to turn it off.

Beta scope

Beta records live on beta infrastructure. Do not put production-sensitive data through the beta environment. Production data handling, storage location commitments, and formal retention terms will be documented at general availability.