Usage And Spend

Usage and spend views show how model traffic behaves across projects, keys, and registered models. Use them to understand cost before changing routing policy.

What Olyx Measures

Olyx tracks:

• total requests
• failed requests
• project spend
• key usage
• model spend
• latency trends
• replay deltas
• infrastructure and model-cost split where available

The precision depends on the metadata available from traces and registered model pricing.

Model Pricing

Cost intelligence depends on the model registry. If a model entry has missing or placeholder pricing, usage views may understate cost.

Before relying on spend reports:

1. Register every model used by the project.
2. Enter input and output token prices.
3. Confirm completed traces include model identifiers.
4. Review one trace detail page to verify cost is non-empty.

Project Spend

Project spend helps you answer:

• Which project is driving model cost?
• Did a routing change increase spend?
• Are staging workloads accidentally using production models?
• Are private deployment paths being compared fairly?

Use separate projects for staging and production so the monthly view remains readable.

Key Usage

Key usage helps you locate traffic sources.

Use one key per service or environment:

support-api-prod
support-api-staging
replay-worker-prod

If a key suddenly spikes, rotate or disable it and inspect recent traces.

Spend Limits

Spend limits are managed in the dashboard where available. Treat them as operational controls, not as the only defense against unexpected provider bills.

Recommended pattern:

1. Set project budgets conservatively during evaluation.
2. Watch trace volume and model spend.
3. Use routing to move simple work to lower-cost paths.
4. Review replay comparisons before changing production routes.

Cost Investigation Workflow

When spend changes unexpectedly:

1. Open Usage to identify the project or key.
2. Open Spend to inspect model-level cost.
3. Open recent traces for that project.
4. Check selected models and routing decisions.
5. Queue replay comparisons against lower-cost model paths.
6. Update routing only after the comparison is acceptable.

Next Steps

• Use Model Registry to keep cost metadata accurate.
• Use Routing to reduce unnecessary expensive calls.
• Use Traces and Replays to compare changes.