Billable Request Metering
LiteLLM Enterprise pricing is usage-based. Billable request metering reports that usage. The proxy counts successful requests to LLM, MCP, and A2A endpoints. It sends one OpenTelemetry counter to LiteLLM's collector. It authenticates with your deployment's OAuth client ID and secret.
Metering sends request counts with the endpoint category, route, status, model ID, LiteLLM version, and license ID when available. Prompts, responses, and virtual keys never leave the deployment. The gateway sends the license key to the LiteLLM license server only if its offline signature check fails. See Activate your license. Metering runs separately from your OTEL logging and metrics pipeline.
Get a LITELLM_LICENSE, OAuth client ID, and OAuth client secret during LiteLLM onboarding. Contact us if you need these credentials.
These OAuth settings replace the certificate settings from earlier builds. Use an image containing the OAuth metering implementation.
Quick Start
1. Set environment variables
| Variable | Required | Description |
|---|---|---|
LITELLM_BILLING_METRICS_ENDPOINT | Yes | Collector URL. Use https://telemetry.litellm.ai |
LITELLM_BILLING_METRICS_CLIENT_ID | Yes | OAuth client ID from LiteLLM onboarding |
LITELLM_BILLING_METRICS_CLIENT_SECRET | Yes | OAuth client secret from your secret manager |
LITELLM_BILLING_METRICS_TOKEN_URL | Yes | HTTPS OAuth token endpoint from onboarding |
LITELLM_BILLING_METRICS_EXPORT_INTERVAL_MS | No | Push interval in milliseconds. Default 60000 |
export LITELLM_BILLING_METRICS_ENDPOINT="https://telemetry.litellm.ai"
export LITELLM_BILLING_METRICS_TOKEN_URL="https://litellm-prod.auth.us-west-1.amazoncognito.com/oauth2/token"
Inject LITELLM_LICENSE, LITELLM_BILLING_METRICS_CLIENT_ID, and LITELLM_BILLING_METRICS_CLIENT_SECRET through your secret manager. The exporter obtains a bearer token with the OAuth client-credentials flow. Both URLs must use HTTPS.
2. Start the proxy
litellm --config config.yaml
3. Verify
The line below is logged at INFO level, so it only shows when the proxy runs with LITELLM_LOG=INFO (or --detailed_debug). Send a request through the proxy, then check the logs for:
Enterprise billing metrics enabled: exporting to https://telemetry.litellm.ai every 60000 ms
If this line is missing, look for a warning naming the misconfigured variable. Invalid startup configuration disables metering. The proxy continues to serve requests if metering fails.
Deploy
- Helm
- Terraform (AWS / GCP)
- Docker
The LiteLLM chart has one billingMetrics block for both deployment modes. Create a Secret named litellm-billing-metrics with client-id and client-secret keys through your secret manager. Then enable metering:
billingMetrics:
enabled: true
secretName: litellm-billing-metrics
| Value | Default | Description |
|---|---|---|
billingMetrics.enabled | false | Enable metering |
billingMetrics.endpoint | https://telemetry.litellm.ai | Collector URL |
billingMetrics.tokenUrl | https://litellm-prod.auth.us-west-1.amazoncognito.com/oauth2/token | OAuth token endpoint |
billingMetrics.secretName | litellm-billing-metrics | Secret with client-id and client-secret |
billingMetrics.exportIntervalMs | "" | Push interval. The proxy defaults to 60000 |
The reference stacks at terraform/litellm/aws (ECS Fargate) and terraform/litellm/gcp (Cloud Run) enable metering when billing_metrics_endpoint is set:
billing_metrics_endpoint = "https://telemetry.litellm.ai"
billing_metrics_client_id = "your-client-id"
Supply TF_VAR_billing_metrics_client_secret through your secret manager. The module stores it in AWS Secrets Manager or Google Secret Manager and injects it into the runtime. Terraform state contains the secret. Restrict access to the state backend. Set billing_metrics_token_url if onboarding provides a different token endpoint.
Use a compatible image containing the OAuth metering implementation, as described in the Docker guide. Add these variables to your existing gateway container:
docker run \
--env-file litellm.env \
-e LITELLM_BILLING_METRICS_ENDPOINT="https://telemetry.litellm.ai" \
-e LITELLM_BILLING_METRICS_TOKEN_URL="https://litellm-prod.auth.us-west-1.amazoncognito.com/oauth2/token" \
-v "$(pwd)/config.yaml:/app/config.yaml:ro" \
-p 127.0.0.1:4000:4000 \
"$LITELLM_IMAGE" \
--config /app/config.yaml
Set LITELLM_IMAGE to your tested image reference. Keep litellm.env private. Include the license, OAuth client ID and secret, master key, and database connection settings from your existing deployment.
What counts as billable
A request counts when it returns a 2xx status on an LLM inference endpoint (chat completions, embeddings, responses, images, audio, and the other inference routes), the MCP transport, or the A2A message/send route. GET reads, management endpoints, health probes, and failed requests do not count.
The count lines up with the successful requests number on the Admin UI usage page.
FAQ
Does this affect proxy performance? No meaningful impact at the default settings. The per-request cost is about 1.6 microseconds.
What if the collector is unreachable? Requests are unaffected. The counter is cumulative, so counts recorded during an outage are included in the next successful export.
What happens on restart? The proxy attempts a final flush on graceful shutdown. A crash or failed flush can lose counts before the proxy exports them.
Air-gapped deployments? Metering needs outbound HTTPS to the collector and OAuth token endpoint. If your deployment cannot reach it, talk to us during onboarding.
