Activity in the workspace
The Activity page shows the latest request records and audit events for your workspace. Request metadata includes the request ID, key ID, model, state, charge, token counts, latency, error code, environment, policy version, and timestamp. It does not include prompt or response text.
The current Activity view is bounded to the most recent 100 requests and 100 audit events. Use the audit export for the full available event history.
Export audit history
Use the dashboard’s audit export. It returns the workspace ID, export time, policy version, chain-check result, and event history. The export is scoped to the signed-in workspace. The dashboard’s session-authenticated endpoints are not a supported external audit API in this release.
Request states
| State | Meaning |
|---|---|
| pending | Reserved and waiting on provider work. |
| completed | Usage settled and answer returned. |
| failed | Known provider rejection; hold released. |
| reconciliation_required | Outcome uncertain or interrupted; hold retained. |
| usage_review | Usage exceeded expected bounds; charge capped and key paused. |
| output_blocked | Synthetic output withheld after a content safeguard; usage charged. |
What the integrity check means
Audit events form an append-only hash chain in the application database. A workspace export can have gaps in the global sequence because other workspaces’ events are excluded. The local verification result is not proof against a sufficiently privileged infrastructure administrator or a substitute for independent audit storage.
Use metadata in support reports
Provide request IDs, model alias, UTC timestamp, HTTP status, and error code. Do not include keys, patient identifiers, prompts, or model responses. Keep your own client request IDs free of PHI too.
Analyze usage in the dashboard
The Activity dashboard offers Overview, Trends and Breakdown views. Overview combines metric cards, top models and API keys, and a chart that switches between spend, requests and tokens. Trends expands the daily charts. Breakdown keeps exact model and key charges visible.
Overview reports support UTC date ranges (up to 366 days), model, API key, status, and request-ID prefix filters. Summary totals cover every matching request. Logs show 50 matching requests per page; the export includes only that page plus aggregate totals, with the applied filters attached.
Inspect a request to see its input and output tokens, saved rates, provider cost, platform fee, charged amount, admission reservation, status, and policy version. An uncertain upstream outcome can retain a reservation; a zero charge does not prove that the provider did no work. Historical rows without saved rates display “Not recorded.”
Reports contain metadata only. No prompt or response text is available for replay. Average completion duration covers completed requests. Time to first token, cache-hit ratios, and reasoning-token breakdowns are not collected.
Archive continuity
The deployed archive worker tracks a durable sequence checkpoint rather than a time window, so a missed run does not age events out of the next export. It verifies the hash chain, writes immutable chunks and advances the checkpoint conditionally. Missing archives, corrupted chains and a checkpoint ahead of a restored database stop the worker and require review.
This is a technical safeguard. Archive protection settings are not a complete legal retention schedule, and provider retention is verified separately.