Export deployment metrics

Scrape Baseten deployment metrics in Prometheus format and send them to your observability provider.

Baseten exposes deployment metrics in Prometheus format at https://app.baseten.co/metrics. Scrape this endpoint to monitor your models in Prometheus, Grafana Cloud, Datadog, New Relic, or another Prometheus-compatible system.

Create a metrics API key

Create a workspace API key for your collector. This key can export metrics but can’t deploy or invoke models.

Baseten CLI

Sign in to the Baseten CLI:

baseten auth login

The CLI stores your login in a local profile. This profile authenticates Baseten CLI commands, but it doesn’t provide credentials to curl, Prometheus, or another collector.

Create the metrics key:

baseten org api-key create \
  --type workspace-export-metrics \
  --name metrics-export

To restrict the key to specific models, repeat --model-id for each model:

baseten org api-key create \
  --type workspace-export-metrics \
  --name metrics-export \
  --model-id <model-id-1> \
  --model-id <model-id-2>

The CLI prints the key once:

Save this key now. It will not be shown again.
abcd1234.ABCDEFGHIJKLMNOPQRSTUVWXYZabcdef01

cURL

Set BASETEN_API_KEY to an existing API key management key, then create the metrics key:

Request

curl --fail --silent --show-error \
  --request POST \
  --header "Authorization: Bearer ${BASETEN_API_KEY}" \
  --header "Content-Type: application/json" \
  --data '{
    "name": "metrics-export",
    "type": "WORKSPACE_EXPORT_METRICS"
  }' \
  https://api.baseten.co/v1/api_keys

To restrict the key to specific models, add model_ids to the request body:

{
  "name": "metrics-export",
  "type": "WORKSPACE_EXPORT_METRICS",
  "model_ids": ["<model-id-1>", "<model-id-2>"]
}

Store the new key in your collector’s secret store. Don’t use a developer’s personal CLI credential for a shared or production collector.

Choose an integration

Scrape the endpoint once per minute. Baseten collects workload metrics every 30 seconds, so a shorter interval increases request volume without providing fresher data.

Use one of the following integration guides:

• Prometheus scrapes the endpoint directly.
• Grafana Cloud connects to the endpoint as a hosted scrape job.
• Datadog receives the metrics through an OpenTelemetry Collector.
• New Relic receives the metrics through an OpenTelemetry Collector.

For metric names, types, and labels, see Supported metrics.

To export application-defined metrics, see Export custom metrics. You can expose a Prometheus endpoint from a custom server or push metrics directly to an upstream provider with OpenTelemetry.

Rate limit

The metrics endpoint accepts six requests per minute per organization. The limit applies across API keys and collectors in the organization. Requests above the limit return HTTP 429 Too Many Requests.

Use a 60-second scrape interval for each collector. If you run multiple collectors, keep their combined request rate below six requests per minute.