AI Cost Management
See where production model spend goes.
Acta attributes recorded model usage to the Workload, Run, supplier, and API key that produced it, while keeping external charges and missing coverage explicit.
Keep the source and limits of each cost visible.
Review recorded model spend without blending Acta charges, external infrastructure, and unknown amounts.
Cross-provider spend
Review recorded model usage across Acta-routed suppliers and customer endpoints.
Customer-endpoint infrastructure may remain externally billed or unknown.
Workload attribution
Connect recorded spend to a stable Workload and the Run that performed the work.
Consistent attribution depends on the Workload identity supplied to Acta.
Cost coverage
Distinguish Acta-recorded, externally billed, incomplete, and unknown cost.
The record states what its available evidence can support.
Spending limits
Set a calendar-month limit on an API key and restrict its model or provider access.
The limit applies to Acta-recorded usage under that key.
Usage history
Review requests, route mix, tokens, latency, cache reporting, and unit cost.
History supports investigation; it does not explain every external cost.
Scenario modeling
Compare current and alternative model costs with explicit rates, retry rates, and acceptance assumptions.
Scenario output is an estimate, not observed savings or proof of equal quality.
Do not mistake a cheaper request for a better production choice.
Use recorded cost alongside supplier history and available evaluation grades. A later production decision still requires evidence about the business result.
- Router
- Supplier economics considered inside the eligible route set
- Evaluations
- Captured output grades with method and context
- Production Record
- Recorded cost, source, coverage, and operational history
Compare one production workflow before changing its model.
The design partner assessment reviews the available cost and quality evidence for the current setup and agreed alternatives.