Process overview. Your real requests and charges appear in Request Explorer.
Service
Published models
Top-up from
One request. Visible economics.
Understand the range
Model, context, reasoning level and output limits produce an estimate. It is never a guaranteed final charge.
estimated/
reserved/
actual/
released
Control plane
Decide before the request is sent.
SmartRoute
Choose a purpose. Inspect the decision.
Ask for an exact model, or use a policy alias. Every candidate is checked against capabilities, availability, limits and price, and the routing reason stays visible.
Policy aliases
smart-code
smart-fast
smart-value
smart-document
smart-premium
An exact model slug is a strict pin. It bypasses alias scoring and never changes to another model.
A candidate is excluded when
Required streaming, tools, vision or reasoning is missing
The provider route is not active or degraded
Requested output exceeds the model limit
Input plus output exceeds the context window
The reservation exceeds the API key per-request budget
Hard limits reject a request before expensive upstream work starts. Reserved Coin and settled spend are tracked separately, and committed in one transaction.
WorkspaceOwner, admin and billing members set the boundary.
ProjectSeparate limits for each environment you run.
API keyRotated keys share one allowance while they overlap.
Each policy can limit
Per request
Daily
Monthly
Protected balance
A hard limit returns HTTP 402 budget_limit_exceeded. A soft limit records a warning. Chanx AI never reroutes a model because of a budget.