Keeping your OpenAI costs low when agents run all day
Agents use tokens every time they judge a job or write a draft. A few settings keep the bill predictable.
When you use your own OpenAI API key, you pay OpenAI directly for the tokens your agents use. That is good for transparency, and it also means a few settings decide whether the monthly figure is small or surprising.
Where the tokens go
An agent spends tokens in three places: judging each job it finds, writing proposals, and drafting replies. Judging usually dominates, because it happens for every job on every run, while proposals and replies happen only for jobs and messages that matter.
The levers that matter most
- How often agents run. Hourly costs about twice as much as every two hours.
- How many jobs are judged per run. Tighten your keywords and filters so fewer weak jobs reach the model.
- How many agents you run. Each one searches and judges on its own.
- Which model each agent uses. A smaller model is often enough for scoring jobs.
Measure, do not guess
Every run in your log shows the tokens it used. After a day or two you will have real numbers to multiply instead of estimates. Compare them with the current prices on OpenAI's pricing page, and watch your usage in the OpenAI usage dashboard.
Set a spending limit
OpenAI lets you manage limits and billing from your account. Create your key in the API keys page and keep a close eye on the first week.