Budgets and rate limits

Cap spend and traffic for your organization, each AI system and each key.

This is in development. Details may change before it is available.

Altrum works out the cost of every request from the tokens the provider reports and a price list that Altrum keeps up to date every day. A model that is not on the list has no price, so its requests are recorded with no cost and do not count against a budget, unless you set your own price for it. Budgets cap that cost, and rate limits cap traffic.

Your own prices

When you pay a different rate (a negotiated price, a model the list does not have), an owner or admin sets it in AI Gateway, Settings, Pricing, with Add a price:

Field
Name	What the price is, for example “Negotiated rate”.
Model	A model name such as gpt-4o, or a prefix ending in *, such as gpt-4o*, for every model that starts with it.
Applies to	Every key, or one key. A key’s own price wins over one for every key.
Provider	Any provider, or one.
Input, Output, Cache read	US dollars per million tokens. Leave one empty to keep the published price for that part. Set at least one.

The price is used for requests from then on, for chat and Responses requests, streamed or not. It changes the cost Altrum records and counts against budgets, never what your provider bills you. Costs already recorded do not change when you edit or remove a price.

Where limits apply

Level	Set in	Applies to
Organization	AI Gateway, Settings, Limits	All gateway traffic in your organization
AI system	The AI system’s AI Gateway tab	All keys of that AI system
Key	Limits when you create or edit the key	That key only

A request must pass every level that applies to it. Its cost counts against each of them.

Budgets

A budget is an amount in US dollars for a period: an hour, a day, a week or a month. When spending reaches the budget, further requests are refused until the period resets. A budget resets a set time after it started, or, when you choose calendar alignment, at midnight UTC for a day, on Monday for a week and on the first of the month for a month. Each organization, AI system and key has at most one budget in the dashboard.

Rate limits

A rate limit caps the number of requests, the number of tokens, or both, over a minute, an hour or a day. Requests over the limit are refused until the window resets. There is no default limit: a new organization has none until it sets one.

What a caller sees

A request refused by a budget says which budget is used up (the key, the AI system or the organization), how much was spent and when it resets. A request refused by a rate limit says which limit was reached. Both come back in the format of the API the caller used. Raise the limit and requests flow again within seconds.

Personal keys

Every personal key starts with the defaults an admin sets under AI Gateway, Settings, Access: a budget and a request limit. Admins can tighten any personal key. A personal key also counts against the organization’s budget.