Free tool
Work out what an AI agent really costs to run around the clock. Enter how often it runs, how much context each turn carries, and your model's prices to see monthly token spend, hosting, and a spend cap worth setting.
Total / month
$89
$50 tokens plus $39 hosting
Token spend / month
$50
Input is about 70 percent of it
Cost per run
$0.73
122 runs a month
Suggested spend cap
$140
Your total plus 50 percent headroom
This agent burns about 13M tokens a month and costs about $89 to keep running. Tokens are the larger half of that bill, so the thing to watch is context size, not the size of the box it runs on.
A retry storm or a task firing more often than intended can easily triple token use. That would take this agent to $190 for the month. A cap of $140 stops it well before that, while still leaving room for a busy week.
The calculator turns your agent's shape into a monthly token count, prices it at the rates you enter, and adds what you pay for hosting. It runs entirely in your browser and nothing you type is sent anywhere.
An agent loop re-sends the conversation so far on every turn, so context grows as a run progresses. That makes input the larger share of most agent bills even though output is priced higher per token. A flat figure per turn is therefore a floor, not a ceiling. Where your provider offers prompt caching the repeated portion gets much cheaper, so check whether the price you enter should reflect that.
Steady usage is rarely what causes a surprise bill. The usual causes are a loop that retries, a task firing more often than intended, or a tool returning far more text than expected. Size the cap against a bad day rather than an average one, and set it somewhere that hitting it stops the agent instead of only warning you.
It depends almost entirely on how often the agent runs and how much context each turn carries, not on uptime. A scheduled agent doing a handful of runs a day usually costs more in model tokens than in hosting. Enter your own numbers above to get a figure for your case.
Qoren runs OpenClaw, Hermes and Codex agents in managed cloud environments, with the spend cap enforced for you. No Docker and no VPS to babysit.
