Overview
Overview shows how much you used the proxy over one day, week, or month: estimated spend, requests, tokens, caching, and which models did the work. It opens first after you sign in.
What is on the page
Time range
The Day, Week, and Month control in the page header sets the range for everything below it.
- Day shows one day, one bar per hour.
- Week shows one Monday to Sunday week, one bar per day.
- Month shows one calendar month, one bar per day.
Choose the selected option again to open a calendar and pick a date, week, or month. Ranges follow your browser's time zone. The refresh button next to the control reloads the numbers.
Metric cards
Three cards sit at the top: Spend, Requests, and Tokens. Each one shows:
- the total for the range,
- a small stacked chart of that metric over the range, one color per model,
- the top four models by that metric, with the rest folded into Others.
Spend is an estimate in USD, not a bill. It prices each request's tokens at the model's public API list price, including requests that went through a subscription. The line under the total says how much of it is priced:
- Estimated from list prices: every request in the range has a price.
- Estimated · covers N of M requests: some models have no known price, so the total is low.
- No priced requests in this range: the total shows
—, not$0.00.
Tokens counts input plus output tokens.
The expand button on a card opens a larger chart and a table with each model's Min, Max, Avg, and Sum per bar over the range. Avg ignores bars where that model had no traffic.
Activity
The card below has four tabs. Its header also shows Errors, the number of failed requests, and Avg latency, the average of the Latency column in Logs.
| Tab | What it shows |
|---|---|
| Tokens | Tokens per bar, stacked by model |
| Requests | Requests per bar, stacked by model |
| Caching | Input tokens per bar, split into Cached and Uncached |
| By model | One row per model with Requests, Errors, Input, Cached, Cache write, Output, Cache rate, and Spend |
Models are always shown by their full provider/model id. A request sent through a model group counts under the model that actually answered. Requests to every account of a provider add up in the same row. To see which account or group served a single request, use Logs.
Some providers do not report token counts for every request. Those requests still count in Requests and Errors but not in token totals. When that happens, Cache rate says how many requests it covers, for example on 40 of 52.
Usage from providers you have since removed is not shown here. Logs still lists those requests.
Common tasks
See what one day cost
- Choose Day.
- Choose Day again to open the calendar.
- Pick the date.
- Read Spend, then expand it to see the cost per model.
Check whether prompt caching works
- Pick a range with traffic.
- Open the Caching tab. A large Cached share means repeated prompts are read from cache.
- Open By model and compare Cache rate between models.
Find the cause of a spike
- Find the tall bar in Tokens or Requests and hover it to see the models in it.
- Go to Logs and filter by that model's provider to see the single requests.
Empty page
No requests yet means no request in the selected range reached the proxy with your keys. Choose Get your API key to create one on the Access page.