Usage analytics
See AI traffic as trends by model, provider, key and team, with tokens, errors and cost over time.
What is Usage analytics?
Usage analytics aggregates your AI traffic into trends by model, provider, API key and project. It shows requests, tokens, cost and errors over time. On Spro the workspace usage page shows these totals, so decisions about routing and models are based on your own traffic rather than estimates.
Single numbers hide what is happening. Analytics show whether cost is rising, which model is driving it, and whether errors came from one provider. That is the information needed before changing a route or model.
Because the totals come from the same request records as the logs, a total can always be traced back to the calls behind it.
Key features
What this feature does for your requests.
- Trends over a chosen date range
- Group by model, provider, API key or project
- Tokens, cost, requests and errors side by side
- Drill down from a total into individual request logs
How it works
4 steps, from setup to the first request.
- 01
Send traffic
Requests through any key in the workspace are counted.
- 02
Open usage
Go to the workspace usage page.
- 03
Choose a range and grouping
Pick the dates and group by model, provider or key.
- 04
Drill into requests
Open the logs behind a number that looks wrong.
Is it the right fit?
Specific situations where this feature helps, and where it does not.
Good fit if
- Comparing models or providers on your own traffic before changing routing
- Finance and engineering need to look at the same numbers
- Checking whether a prompt, model or cache change moved cost or errors
Not a good fit if
- Investigating one failing call. Request logs show that detail
- Judging answer quality. Analytics measure usage, not output quality
- Looking at multi-step agent runs. Tracing is built for that
Usage analytics FAQs
Common questions about Usage analytics.
Route every AI request through SPRO.
One OpenAI-compatible API for 350+ models, with intelligent routing, 29 built-in guardrails, observability, and automatic failover.
