Set up automated model reports
Pin a reporting job to a saved baseline, test the comparison and set reporting expectations.
Open a model in your Prompto workspace, then Overview & reporting. Local checks remain free and offline; account reporting starts on Hobby. Your own CI runner or operating-system scheduler runs the command.
- Choose a saved model lockfile. Its fingerprint and record date identify the exact baseline. Saving a newer version will not change this job's reference.
- Select or create a reporting token. Save its secret as VRAELIS_REPORT_TOKEN in your job's secret store. A new token is shown once.
- Enter the model folder and choose Bash or PowerShell. Select Start test, then run the generated command where the CLI and files are available. It creates a fresh temporary snapshot and submits only after a successful read.
- Refresh the test result. A new server comparison from this token against this exact baseline confirms the reporting connection. Changed means the connection worked and the incoming files differ; review the report before accepting an update.
- Configure change alerts and an expected-report schedule. The schedule expects reports to arrive; it does not run the command for you.
Comparison identity
The report endpoint accepts baseline_id as a query parameter. An explicit ID must be an account-owned, compatible saved lockfile under the same model name. A missing, deleted or incompatible ID is refused without falling back. Result-only submissions cannot be independently compared and cannot satisfy this setup test.
Existing jobs that omit baseline_id continue comparing with the latest saved baseline by model name when it is compatible. Reports now record the actual baseline ID, fingerprint and selection mode. Older reports have no recorded baseline identity. Deleting a baseline later does not remove that identity from its reports.
Reporting health and failures
Received at means the server received the report, not that it witnessed a fresh disk read. The supplied command reads files on each run, stops on CLI errors, surfaces HTTP errors and removes its temporary snapshot. Store the command in your own scheduler for recurring reports.
Expected-report schedules cover the exact model name across jobs and baselines in your account. Any accepted report or saved check under that name can satisfy the schedule. Changing the baseline does not create a separate schedule.
To move a job to a newer baseline, open that saved version, generate the replacement command and test it before changing the scheduled job. Keep the previous baseline available if you need to compare against it later.
Last updated: November 12, 2026