Production API health.
The badge below is fetched directly from the Modal-hosted FastAPI /health endpoint by your browser. There is no caching layer between you and the live response. If the request fails, the badge says so.
fourwheels2512--crma-finetune-fastapi-app.modal.run/health
What we monitor.
Five components carry traffic for ModelBrew. The dates below are the most recent date a human or smoke test verified the component end-to-end — not a claim of zero downtime in between.
fourwheels2512--crma-finetune-fastapi-app.modal.runGET /health returns 200 with {"status":"healthy"}. Live-fetched on this page every 60 s.app.modelbrew.aimodelbrew.ai200 from Cloudflare Pages with their canonical content.Gemini 2.5 FlashStripe webhookPOST /stripe/webhook verifies signatures, the credit-grant path is idempotent, and a synthetic checkout-session event credits the test account.What happens when something breaks.
Incident response policy
Detection — The live badge on this page is the first public signal. Internal smoke checks and Modal's own platform alerts are the second. The founder is paged via email on the on-call address.
Notification — For incidents that impact billing or a user's ability to launch a paid run, we email each affected user from modelbrewai@gmail.com within 1 business day of detection — whether or not the incident is fully resolved by that point.
Auto-refund — If a fine-tuning or continual-learning run gets stuck (no heartbeat past the active-job window), the backend auto-refunds the held credits within ~90 minutes. The refund path is in backend/server.py (_auto_refund) and is logged with a correlation ID so the operator can reconcile against Stripe.
Postmortem — For incidents lasting more than 1 hour or involving data exposure, we publish a short postmortem (cause, blast radius, fix, prevention) on the blog within 7 days.
Subscribe to incident notifications.
For now, the founder is on call.
ModelBrew is an early-stage startup with a small founding team. There is no Tier 1 SRE rotation; a founder is the on-call engineer and reads every email at modelbrewai@gmail.com.
If you want to be notified about incidents that affect a run you launched, just reply to your most recent invoice email and ask to be on the incident list — we'll add you and email you whenever a billing-impacting issue happens.
We are evaluating Better Uptime for a full status page with a self-serve subscriber list and historical uptime numbers. When that ships, this widget will be replaced by the Better Uptime embed and you'll be able to subscribe yourself in one click.