Monitoring & Logs
BRICQS collects real-time metrics and logs for every deployment automatically — no instrumentation required. Metrics are collected by the BRICQS monitoring layer; logs stream directly from your container's stdout/stderr.
Available metrics
CpuPercentageAverage CPU utilization across all replicas, sampled every 5 minutes.MemoryPercentageWorking set memory as a percentage of the allocated memory limit.RequestsTotal HTTP requests received by the deployment in the interval.ResponseTimeAverage end-to-end response time in milliseconds, including model inference.ReplicasNumber of running replicas at the sample point.GpuUtilizationPercentageGPU compute utilization (GPU-enabled deployments only).RestartCountContainer restart count. Elevated values may indicate OOM or crash loops.Viewing metrics
Metrics are visible in the Deployments → [select deployment] → Metrics tab. Charts default to a 6-hour window; you can switch to 1h, 24h, or 7d.
You can also query metrics directly from the API:
bashcurl https://api.bricqsai.com/v1/deployments/<id>/metrics?hours=6 \
-H "Authorization: Bearer <api-key>"Container logs
All stdout/stderr output from your deployment containers is captured and searchable:
bash# Tail live logs
bricqs logs <deployment-id> --follow
# Filter by time range
bricqs logs <deployment-id> --since 1h
# Search logs
bricqs logs <deployment-id> --grep "ERROR"Logs are retained for 30 days. For longer retention, configure log export to your own BRICQS Storage bucket in the project settings.
Runtime logs
Runtime logs capture LLM API call events — prompts, completions, token counts, latencies, and model IDs — emitted by the BRICQS runtime layer (not the container directly). View them in Runtime Logs in the console, or query via:
bashcurl https://api.bricqsai.com/v1/runtime-logs \
-H "Authorization: Bearer <api-key>" \
-G --data-urlencode "project_id=<project-id>" \
--data-urlencode "limit=100"Deployment Intelligence alerts
BRICQS automatically surfaces operational recommendations when your deployments show anomalous patterns:
- Idle deployment — no traffic in the last 24 hours; consider stopping to avoid billing.
- Repeated failures — ≥3 consecutive container restarts detected; check logs for root cause.
- High latency — average response time exceeds 3 000 ms in the last 24 hours.
- Budget warning — GPU-hour spend is on pace to exceed the project's monthly limit.