Your portfolio, running itself.
Loope is a small, auditable staff for everything you've shipped. It watches every product, names what broke, proposes the fix, and writes you the Monday brief. Every action is typed, costed, and on the books.
Bring your products. The first sweep runs within ten minutes.
Sonave is down: HTTP 500 after 3 attempts.
- Cause
- The application is crashing, not the infrastructure. http-5xx
- Fix
- Check the function logs for /api/auth/guest.
One loop, closed every ten minutes. It stops once, for you.
Each stage is a typed hand-off. Nothing skips a stage, and every pass through the loop leaves a row in the ledger.
- 01
Watch
Sentinel probes every production URL. A bad observation is confirmed with retries before it becomes an incident.
GET holdouthq.com/ 200 · 145 ms - 02
Diagnose
Not “down”: the cause. Stale DNS, wrong project on the domain, the app throwing 5xx, latency over budget.
http-5xx the application is crashing, not the infrastructure - 03
Propose
When a fix is executable it is typed: redeploy, attach domain. “Investigate” is a recommendation, never an action.
attach-domain cineengine.com to cineengine - 04
Approveyour yes
Nothing production-facing runs without you. One click from the queue, and the decision is recorded with your name on it.
pending approve and run · reject - 05
Execute
The Operator runs it through your own credentials and records the outcome, including failure.
vercel: attach cineengine.com to cineengine - 06
Report
Monday, 13:00 UTC. The Analyst writes the brief from the week’s numbers and ends it with decisions, not charts.
brief three products are down with open incidents; the rest of the portfolio is healthy
Six roles, in order. Three are live.
Volume is automated. Anything that speaks for you waits for you.
| Role | Job | Cadence | Gate | Status |
|---|---|---|---|---|
| Sentinel | Watches every product, confirms before alarming, names the cause. Live. | every 10 minutes | Unattended | Live |
| Operator | Executes approved fixes through your credentials and records the outcome. Live. | on your yes | Gated | Live |
| Analyst | Writes the Monday brief from the week’s numbers. Labelled when it falls back to the rule-based writer. Live. | Mondays, 13:00 UTC | Unattended | Live |
| Scout | Runs outreach. Next. | — | Gated | Next |
| Concierge | Answers support. After Scout. | — | Gated | After Scout |
| Scribe | Writes content. Later. | — | Permanently gated | Later |
Nothing runs off the books.
Every agent run, clean or failed, is one ledger row: duration, model calls, cost. Accounting was built on day one, not retrofitted.
Nothing outward without your yes.
Production-mutating actions wait in a queue. Agents handle volume; you keep outcomes.
Failure is data.
A wrong heal proposal taught the system a new diagnosis class. Fallbacks are labelled, never disguised.
Your keys stay yours.
Workspace credentials are encrypted at rest, write-only, and never leave your tenant.
loope · loope.dev/api/health
probed every 10 minutes by its own Sentinel
in production since 4 Aug 2026
the incidents on this page are its own findings