What it costs to run an AI agent

The most common question is not "can it?" but "what will it cost per month". The answer has three parts, each behaving differently.

1. The environment (container)

An agent needs a process, memory and disk. While it runs you pay for hours; while it sleeps you pay nothing. Sleeping when idle is therefore the strongest lever on price — often stronger than any discount.

For scale: an agent working two hours a day uses roughly 6% of the month. The rest is sleep.

2. The model

The second part is the brain. Tokens decide the price — how much text the agent reads and writes. Long documents cost more than chat.

Practical tip: have the agent extract only what matters first (prices from a page, rows from a PDF) and summarise afterwards. It often halves the bill.

3. Storage

Files are cheap — a few crowns per gigabyte per month. But an untidy workspace grows: downloaded pages, logs, working copies. Clean it occasionally.

When it pays off

When it replaces repeated work. If a weekly digest takes you three hours, payback is quick. Twice a month? Doing it by hand is cheaper.

Beta

MultiBrains is in free beta right now. Pricing starts after the service is polished, announced at least a month ahead. Numbers are in pricing.