Control AI spend before it happens
Set budgets for the organization, teams, routers, people and models. Get alerts before a limit, block at it, and know what every customer and feature costs.
Budgets for the organization, routers and tags
A budget covers the whole organization, a list of up to 200 routers, or every router carrying one of up to 50 tags. Each budget sets a total cap, a cap on each router and caps on single models, with its own Notify and Block rules.
Caps for every team, person and model
Scope a budget to the organization, a team, a tag, a router or a person. Each one takes a Total cap, a cap on every line and caps on single models, across the whole budget or inside one line, with Notify and Block rules. Policies create the same budget for every team or person they match.
Soft and hard limits
Every budget can carry soft and hard limits. A soft limit alerts by email or signed webhook when spend reaches a percentage of a cap, or is projected to. A hard limit blocks: callers get HTTP 429 budget_exhausted with Retry-After until the period resets.
HTTP/1.1 429 Too Many Requests
Retry-After: 259200
{
"error": {
"message": "budget bdg_7Hq2Lm4NwQ is exhausted for this period; it resets at 2026-11-01T00:00:00Z",
"type": "budget_exhausted",
"code": "budget_exhausted"
}
} Overrides and requests for more
Raise one limit for the rest of the period without editing the budget. Employees ask for more in the Portal, request approvers approve at the amount they choose, and every override ends when the period resets.
Emailed when an employee asks for a higher limit in the Portal.
Every price, every region, your negotiated rates
One model can carry a dozen prices: by vendor, region and variant, by service tier, for cache writes that live five minutes or an hour, for cache reads, long prompts and web searches. Multi-Router prices each request at the exact rates that served it. Set your own negotiated rates and budgets, caps and alerts use them instead of list prices.
Set your negotiated rates once. Every request is priced at them, so budgets, caps, alerts, Traffic and Detailed usage count what you actually pay.
Know what each customer and feature costs you
Create a router for each customer and feature through the API, tagged with customer and feature, and send a session id and a correlation id with every call. Spend adds up from each request to the workflow run, the feature and the customer, so you know the AI cost behind every account and what each feature costs to run.
Put a budget on your AI traffic
Create a budget, set its caps and rules, and see spend against it on every request.