Skip to main content
Scheduled runs let a team re-execute the same benchmark on a fixed cadence (daily, weekly) without any human trigger. That is how you catch regressions caused by upstream changes, like a model provider shipping a new point release or a tool dependency changing behavior, that would otherwise only surface the next time a human happened to run the benchmark.
Native scheduled runs are on the roadmap. In the meantime, use any cron-based workflow runner (GitHub Actions, Cloud Scheduler, Temporal, BullMQ, etc.) to call POST /v1/benchmark-runs on a cadence with your Solver key.

​
Interim: GitHub Actions

GitHub Actions exposes a schedule trigger that uses cron syntax. Reuse the workflow from Run a Benchmark in GitHub Actions and add a schedule trigger alongside pull_request: 
on:
  pull_request:
  workflow_dispatch:
  schedule:
    - cron: "0 8 * * 1" # every Monday at 08:00 UTC
Every scheduled run goes through the same code path as a pull-request run, so the score, verdict, and evidence are directly comparable across cadences. Pair the scheduled run with a webhook subscription on benchmark_run.completed to page the on-call when the score drops below threshold.

​
Planned Behavior

The planned native feature will let you, from the dashboard or API:
  • Pin a Solver, benchmark slug@version, and cadence (daily / weekly / cron).
  • Automatically compare each scheduled run to the previous run and surface regressions.
  • Optionally hold the run for manual review instead of auto-completing.
Until that ships, the cron + webhook pattern above covers the same shape of workflow.

​
Next Steps

Webhooks

Get paged when a scheduled run completes with a dropped score.

GitHub Actions

The current recommended way to schedule Verial runs.