Skip to main content

​
Credits & metering

Credits are Starfire AI’s internal unit for measuring eligible compute and feature usage across the platform. The metering system is designed to support both interactive web usage and developer API workloads while preserving enough detail to understand where usage came from.

​
Usage categories

Usage can be attributed to categories such as:
  • chat generation
  • reasoning
  • research
  • search
  • FORGE builds
  • embeddings
  • reranking
  • developer API requests
  • other metered tools or capabilities
The exact credit formula is controlled by the active model, tool, plan, and pricing configuration.

​
Included, bonus, and purchased credits

A billing context can track more than one source of available credits, such as plan-included allowance, administrative bonus credits, or separately purchased credits where supported. A useful account view separates those sources instead of presenting only one opaque balance.

​
Hard limits

Starfire can enforce usage boundaries such as:
  • monthly allowance
  • daily or monthly ceilings
  • feature-specific budgets
  • agent/workflow budgets
  • API quotas
  • organization-level shared limits
A hard ceiling prevents new metered work after the configured threshold is reached, even if the underlying provider would technically accept another request.

​
Personal vs organization metering

Usage should be attributed to the context that owns the work. A request made inside an organization-owned project can draw from organization entitlements and credits rather than the member’s unrelated personal allowance.

​
Web and API usage

Starfire’s architecture uses a shared metering foundation while still allowing reporting to distinguish interactive web usage from Developer Platform usage.

​
Administrator visibility

Authorized administrators can inspect usage totals, category breakdowns, allowances, and limits. Credit adjustments should be auditable, especially actions such as adding credits, changing allowances, resetting a period, or setting a hard ceiling.
The documentation intentionally does not publish a fixed credits-per-token formula because model costs and Starfire pricing multipliers can change independently from the docs.