Feature

Usage Dashboard

Every request updates the dashboard within seconds — spend, tokens, latency, and errors, broken down by key and by model. The same numbers are available through a usage API if you want them in your own reporting instead.

How it works

Every response is logged

Tokens in, tokens out, model, latency, and cost are recorded the moment a response is returned, tagged with the API key that made the request.

Aggregated in real time

The dashboard rolls that data up by key, model, and time window with no batch delay, so a spend spike shows up immediately.

Queryable the same way

Anything visible in the dashboard is also available through the usage API, so you can pipe it into your own tools or alerts.

usage-query.ts

const usage = await aniron.usage.query({
  start: '2026-09-01',
  end: '2026-09-10',
  group_by: ['api_key', 'model'],
});

// usage.rows[0] =>
// { api_key: 'prod-web', model: 'aniron/llama-3.3-70b',
//   requests: 48213, tokens: 9482113, spend_usd: 12.44 }

< 5s Data freshness

Median delay between a response being billed and it showing up in the dashboard and usage API.

When to use the usage dashboard

Catch a runaway integration

Spot a key burning through its budget hours before the recurring billing cycle would have surfaced it.

Compare model cost and quality

See spend and latency side by side across the models you route to, so you can decide which one is actually worth the price.

Attribute cost to a customer or team

Filter by API key to see exactly what a given customer, feature, or internal team is spending, without building your own metering.

Diagnose a provider outage

Watch error rate and latency per model climb in real time so you know a provider is degrading before support tickets do.

Frequently asked questions

How current is the data in the dashboard?

Spend and token counts update within a few seconds of a response being returned. There is no end-of-day batch job or billing delay.

Can I see usage broken down by model and by key at the same time?

Yes. The dashboard and the underlying usage API both support grouping by any combination of API key, model, and time bucket.

Does the dashboard show failed requests?

Yes. Failed requests are shown separately from billed usage, since a failed request is never charged. You can see error rate per key and per model to spot a degrading provider.

Can I export usage data instead of viewing it in the dashboard?

Yes. The same data backing the dashboard is available through the usage API as JSON, or as a CSV export, so you can pull it into your own billing or analytics pipeline.

Can I set up alerts on usage instead of checking the dashboard?

Yes. You can configure a webhook or email alert on thresholds like total spend, spend on a specific key, or error rate over a rolling window.

Related resources