All systems operational. Updated continuously.
Every public AgentLoka service is responding within normal bounds. Subscribe to be notified the moment anything changes.
Per-service health.
Every day. Every service.
Newest day on the right. Green = operational. Amber = a minor incident that didn't fail customer runs. Red = a true outage. Hover any day for the date.
The last 90 days,
Run worker pool exhausted in EU-Frankfurt
Between 14:02–14:38 UTC, runs queued in EU-Frankfurt experienced p95 start latency of 38s (normal: 1.2s) due to a slow-leaking worker that survived health checks. A single misbehaving customer agent in an infinite tool-call loop monopolised four workers; our concurrency limiter only kicked in on the fifth. We've shipped a stricter per-agent worker cap (max 2) and added a synthetic 'is this worker doing real work' check that runs every 60s. No data loss, no failed runs. Every queued run completed once the pool was rebalanced.
Webhook delivery delayed for 8 minutes
A noisy neighbour on the shared notifications cluster triggered a self-healing restart that took 8 minutes longer than expected. All queued webhooks delivered with retry; no events were dropped. Moved webhooks to a dedicated cluster on 2026-05-17.
OAuth refresh failures for one upstream provider
An upstream API rotated its OAuth signing key without notice. Existing access tokens kept working; refresh attempts failed silently. Detected via our synthetic check at 09:14 UTC; mitigated by 09:41 by switching to the new key. ~430 buyer connections required a one-click re-auth.