What does an AI agent cost, and what are you paying for?
Pricing pages compare badly because they measure different things. What follows is the full shape of the cost, so you can compare two quotes that look nothing alike.
Questions and answers
What are the parts of the bill?
Three. The platform — the agent definitions, the runs, the audit trail. The model tokens, which scale with how much text goes in and out. And your own time setting it up and checking the output early on. Vendors quote the first, argue about the second, and never mention the third, which is often the largest in month one.
Why do model costs vary so much between tasks?
Because you pay for text, not for time. Reading a forty-page contract costs many times more than triaging an email, whatever the task is called. A task with a small input and a small output is cheap even if it feels sophisticated.
Is a subscription or a one-off purchase better?
A subscription if you will use it monthly and want it to keep improving; a one-off if it solves a bounded problem you will not revisit. The question that decides it is whether the work recurs, not whether the price looks smaller. Current plans are on the pricing page.
Can I try one without paying?
You can read every agent in full without paying anything or connecting anything. Running one needs your own provider key: a free account gets one run a month, and the model request goes to your account at your provider's price. PROMIVO charges for the platform and the agents, never for the tokens.
What does a custom agent cost compared with one off the shelf?
More, because someone has to define it, and the honest reason to pay it is that your process is genuinely not the standard one. If a catalogue agent gets you eighty per cent of the way, the last twenty is usually cheaper to absorb than to specify. See the custom agents page for how briefs are scoped.
How do I avoid a surprise on the bill?
Watch input size rather than run count. Ten runs over long documents cost more than a thousand over short ones. Set a limit before you scale a task up, and look at the usage figures after the first week rather than at the end of the month.