Budgets & Usage Limits

V2 console

This card lives in Policy Engine > LLM Gateway at web.quilr.ai/policy. Edits join the shared draft and take effect once you review and publish a revision.

Cap spend in USD, or cap requests and tokens, over a reset period. Count it shared across everyone in scope, or separately per person, application, key, model or credential.

How budgets work

• Every matching budget applies. Budgets do not resolve by priority. A request must fit inside all of them.
• Requests count at the start; tokens and spend count at completion. One in-flight request can carry a counter past its limit.
• Once a budget is reached, later matching requests get HTTP 429 quota_exceeded until the period resets.
• Spend needs prices. A spend budget prices provider-reported usage with the rates in Model pricing. With the Policy Engine on, this card is the only price source (prices set under Settings > Models are not used): a model with no rate under a spend budget is rejected with HTTP 503 quota_price_unavailable.
• With no budget configured, spend and usage are recorded but never capped.

Sections

Section	What it does	Applies on	Empty state
Budgets	Spend or usage counters with a limit and reset window. Add budget.	assistants, bedrock, chat, embeddings, realtime, rerank, responses, stt, text, tts, vertex	Spend and usage are recorded but never capped.
Model pricing	USD per 1 million input and output tokens, per provider and model. Add pricing.	Same	A spend budget cannot price any model until a rate is set.

For a budget keyed on request metadata or an API key, use Add configuration at the bottom of the card and finish in the full editor.

Budget settings

Setting	Options	Default	Notes
Applies to	Everyone, People, Smart group, Application, App tag, Requested model, Provider, API surface, Environment, Source network, Except...	Everyone	Or pick an application directly. Decides which requests count toward the budget.
What to count	Spend (USD), Requests, Input tokens, Output tokens, Total tokens	Spend (USD)
Limit	A number above zero	-	USD for spend (for example 2500), a count otherwise (for example 1000000 tokens).
Counted	Shared, or Separately for each of: Tenant ID, User ID, User email, Application key ID, Application key, Application method key, Application method type, Actual provider, Actual provider label, Actual model	Shared	Pick several fields for one counter per combination, for example each person + each model. Requests without a user email do not count toward a per-person budget.
Resets	Rolling minute, hour, day, week, month or year; Calendar day, week, month or year; Lifetime (no reset)	Calendar month	Rolling looks back over the last period. Calendar resets on the boundary.
Timezone	IANA timezone	UTC	Calendar periods only.
Severity	Not set, Very low, Low, Medium, High, Critical, Very critical	Not set	Reported only. Never changes the outcome.
More options	Open in full editor	-	Priority, extra rules, metadata and content conditions, raw QuilrQL.

Each budget gets a generated ID that its usage is tracked against. See The Budget ID matters before changing a published budget's measure, period or grouping.

Price every reachable model first

For a Spend (USD) budget, Model pricing must cover every model the budget can reach, including fallback models from routing, for the whole budget scope. Coverage is checked when you add the budget and again before publishing.

Budget groups

A budget group is the set of requests that share one counter. Two settings shape it:

You want	Applies to	Counted
One pool for a whole app	Application	Shared
An allowance per person in an app	Application	Separately for each: User email
An allowance per person per model	Everyone	Separately for each: User email + Actual model
A cap per API key	Everyone	Separately for each: Application key ID
A team cap on top of personal caps	Two budgets on the same scope	One Shared, one per User email

Because every matching budget applies, the tightest one stops the request first.

Model pricing settings

The Add pricing dialog first asks who the rates apply to (same scope chips as a budget), then opens the editor to set the rates.

Setting	Notes
Provider	The provider the rate is for.
Provider credential	Optional. Price one credential differently from the provider's other credentials.
Model	The model id as reported by the provider.
Input cost (USD per 1M tokens)	Required for spend budgets.
Output cost (USD per 1M tokens)	Required for spend budgets.

Rates are scoped like any other setting. The simplest setup is one Everyone configuration listing every model you use.

Examples

support_copilot_budgetsrequest

runs on request·priority 500

WhenApplicationisSupport Copilot

Then

Budgets→2 budgets

Budget 1 Spend (USD), 500, Calendar month, UTC, separately for each User emailBudget 2 Total tokens, 10,000,000, Rolling week, shared

Model pricing→2 models

gpt-4.1 input $2.00, output $8.00 per 1M tokensgpt-4.1-mini input $0.40, output $1.60 per 1M tokens

Each person gets 500 USD a month, and the whole app shares 10M tokens a rolling week. A person under their own budget is still stopped once the team cap is reached.

interns_daily_requestsrequest

runs on request·priority 600

WhenSmart groupsincludes (ignoring case)Interns

Then

Budgets→Requests, 200, Calendar day, separately for each User email

A usage allowance needs no prices: each intern may send 200 requests a day.

Scoping and precedence

• Budgets add up; they never override each other.
• Every matching Model pricing configuration contributes its rates. Avoid pricing the same model differently in overlapping scopes.
• Budgets cannot be scoped by Prompt complexity, Prompt text or Tool.

Legacy app setting

Classic app settings have no USD budget. App-wide request and token limits are in Rate and Token Limits.