Cheap LLM API with a spend cap

Who this page is for

Developers and small teams searching for a cheap LLM API with a hard spend cap usually want two things: a low price per token and a bill that stops at a number they chose. This page explains exactly what Avenbridge offers on both, so you can decide whether it fits.

One point first: Avenbridge is a hosted knowledge MCP, not a general-purpose model API. A business forwards its emails and documents, confirms its answers, and its buyers ask them from their own AI through MCP. The prices below are what the business pays for the AI work Avenbridge does on its behalf.

The price

Hosted by Avenbridge: $10 per million tokens. New customers: the first 10 million tokens at $8 per million. There is no monthly fee and no per-seat fee.

Billing is monthly in arrears for the previous calendar month’s usage, in US dollars. Prices exclude any taxes that apply. Private deployment, with your own model and your own servers, is priced by quote.

The spend cap

Every account has a default monthly cap of $300. You can set a lower cap at any time. Once the cap is reached, usage stops being charged for that month, so the largest hosted bill for a month is the cap you set.

A raised cap takes effect only after the owner confirms it from the email sent to their work address, so a cap change needs the account owner.

What counts as tokens

Tokens are the units the model reads and writes when Avenbridge does work for you: reading an incoming email and the material it needs, drafting a reply, noting what a reply promised, and similar tasks. Input and output tokens of that work are counted.

Questions a customer asks in their own AI use no Avenbridge tokens: their AI answers from your confirmed material, and that model is the customer’s. As a guide, answering one customer email takes about 6,400 tokens, roughly 6 cents; noting what a reply promised takes about 4,000. Each bill shows what the same work would cost you in ChatGPT or Claude.

A worked example

At $10 per million tokens, $300 covers 30 million tokens in a month, roughly 4,700 email replies at about 6,400 tokens each. A new customer’s first 10 million tokens cost $80 instead of $100. If you set a cap of $50, the month’s hosted bill stops at $50.

What is included

The customer connector with no usage limits, a public confirmed-facts page, an assistant email address, cases, promise reminders and weekly reports, unlimited colleagues, and setup by email, step by step. Cancel anytime; usage already incurred is billed on the next monthly invoice.

See it first

The try link adds Avenbridge itself to Claude with a personal connector address, so you can ask about the product, plans and security before you sign up.

Last updated: 31 October 2026