Skip to main content
Email developers@reflections.ai for help with your integration: errors you cannot explain, decisions you think are wrong, rate limit increases, and anything this documentation does not answer. The Talent API serves the fictional sandbox and three live account/license reads. Start with the quickstart for the test journey. Hosted conformance runs are not available. No support response time or SLA is promised on this page.

​
Before you write

  • Check status.reflections.ai for a current incident.
  • Look up the error code in Errors, or the decision’s reasons[].code in Reason codes.
  • For a sandbox problem, reproduce it with a supported endpoint and fictional inputs. For live reads, send request and object IDs without copying private license terms.

​
What to send

Request IDs help trace a report. Use fictional prompts and artifacts in the sandbox and send IDs instead of their contents. The documented production privacy policy remains: for a deny or an escalation, the prompt text is kept for 30 days for review and disputes, so there is no need to send it again, unless the check sent prompt_handling.retention: "none". For an allow, the text was deleted once the check was decided; the receipt keeps its digest.

​
What never to send

Never send a secret key (rk_live_… or rk_test_…), a webhook secret (whsec_…), or an end user’s personal data: names, email addresses, phone numbers, or prompts and outputs that contain them. Send request and object IDs instead. If you sent a secret by mistake, revoke or rotate it at once.
Do not send whole authorization or session tokens. The check or session ID identifies them.

​
Rate limit increases

The proposed limits per account for a future live service are 200 checks per second, 50 generations per second, 1,000 concurrent sessions and 50 requests per second for everything else. They are not current sandbox capacity guarantees or an offer of live access. For a sandbox limit question, write with your account ID, the limit you need, your current peak, and when you expect it. See Rate limits.

​
Decisions you disagree with

Sandbox decisions are simulated. Report an unexpected fixture result with its request ID. For future live integrations, support can explain a decision and fix an evaluation defect. It cannot change a license, widen it, or decide an escalation: those belong to the talent, their representatives, and the reviewer named on the escalation. To ask for a wider license, use the license_request or license_change remedy with POST /v1/license_requests.

​
Security reports

Report vulnerabilities to security@reflections.ai, not to developer support. See Security.

Monitoring

Request IDs, alerts and the status page.

Errors

Every error code and what it means.