FinOps
Model spend by provider, workflow and step, computed from every journalled model call at the configured price. Replay the same workload on another model, see which steps are candidates for a smaller or local model, and route them without touching a spec.
Page window
Workflow
Model
Served
Spendreading the ledger…
No priced model calls in this window.
| Model | Served | Calls | Tokens | Avg latency | $/call | Cost | Share |
|---|---|---|---|---|---|---|---|
| No model calls in this window. | |||||||
No workflow spend in this window.
| Workflow › step | Pattern | Calls | Avg in / out | Avg latency | Tools | Model | Cost | Share |
|---|---|---|---|---|---|---|---|---|
| No steps made model calls in this window. | ||||||||
| Step | Verdict | Now | Hand off to | Cost now | Projected | Saving | Why | |
|---|---|---|---|---|---|---|---|---|
| No steps to assess in this window. | ||||||||
Projected figures replay this window’s journalled tokens at the target’s price; on a local provider they price the hardware hour at the provider’s rate with a 1.5× latency allowance. Evidence comes from the latest Compare models batch for the workflow, which measures the whole workflow, not one step.
Workload: 0 calls in 0 runs0 in · 0 out · 0ms of model time
Actual$0.0030d as journalled
Projected—pick a target
Difference—
Every offered model on this workload
No model comparison on record yet. Run Compare models on a workflow’s eval set to get evidence here.
| Workflow › step | Pattern | Spec says | Effective model | Served | Override | Calls · 30d | Cost · 30d |
|---|---|---|---|---|---|---|---|
| No workflows in the registry. | |||||||
Journalled · 30d$0.00
Run rate · month$0.00
Cloud billed · month to datebilling connector not configured
Cloud billed · 7d—
perRuns are refused once the rolling period’s spend reaches it.