URL: /geist/guides/monitoring

---
title: Monitoring
description: Latency, error rate, and log streaming, with no agent to install.
icon: activity
---

Every Halo deployment ships with telemetry on by default. You see traffic at the edge, errors per route, and structured logs from your runtime — all in the dashboard or via the CLI.

## Live logs

```bash
halo logs --tail
```

Streams over a long-poll WebSocket. Filter:

```bash
halo logs --tail --route /api/checkout --status '>=500'
```

## Metrics

The dashboard exposes:

- **p50 / p95 / p99 latency** — per route, per region.
- **Requests per second** — broken down by status class.
- **Error rate** — 4xx and 5xx as separate series.
- **Cold start rate** — for edge functions, the share of invocations hitting a fresh isolate.

All metrics are pulled at 10s resolution and retained for 30 days.

## Alerts

```bash
halo alerts add \
  --name "checkout 5xx" \
  --route /api/checkout \
  --metric error_rate \
  --threshold 0.02 \
  --window 5m \
  --notify slack:#oncall
```

Notify channels: `slack`, `email`, `webhook`, `pagerduty`.

## Exporting to your stack

| Stack | Mechanism |
|-------|-----------|
| Datadog | Native integration — paste the API key in dashboard settings |
| Grafana / Prometheus | OTLP push endpoint at `https://otlp.halo.app/v1/metrics` |
| Anything | Webhook on every alert + log drain to S3 / GCS |

<Tip>
  Need traces? Halo emits OpenTelemetry spans for every edge invocation. Configure the OTLP exporter URL and you'll see your runtime's spans alongside the edge spans Halo emits.
</Tip>