Metering
Metering
Metering listens to the laravel/ai lifecycle rather than provider SDKs. That keeps the integration stable across providers and lets every call become an AiCallEnvelope.
sequenceDiagram
participant App
participant AI as laravel/ai
participant F as FinOps listener
participant P as Provider
participant L as Ledger
App->>AI: prompt, embed, stream
AI->>F: pre-flight event
F-->>AI: allow or block
AI->>P: provider request
P-->>AI: response and usage
AI->>F: post-flight event
F->>L: append usage record
Captured context
| Field | Source |
|---|---|
| provider and model | laravel/ai event metadata |
| tokens | provider usage or estimator fallback |
| cost | cost resolution cascade |
| tenant, user, cost center | configured resolver or trace context |
| trace and agent step | TraceContext |
| purpose | explicit metadata |
Agent step attribution
Wrap each agent step in TraceContext::within() so one run can be reported as a cost flame graph by trace_id.
Beyond the run total
Metering records one priced row per call. For where inside a run that cost went, how long each
tool took, what a run that failed had already spent, and which run called which — see
Run Observability, which needs laravel/ai ^0.11.
Last updated:
Edit this page
Recent commits
