The honest answer
The agent itself is free and open source. Your monthly cost is almost entirely model tokens, and you control that with budgets and model choice.
The three cost buckets
| Bucket | Typical cost | Notes |
|---|---|---|
| Software | $0 | MIT licensed, no per-seat fee |
| Model tokens | A few cents to a few dollars | Depends on model and usage |
| Local (Ollama) | $0 | Runs on your own hardware |
Why agents can get expensive
An agent makes many model calls per task, not one. A chat wrapper costs one turn; an agent costs a plan, a few tool calls, a reflection, and an answer. Budgets exist because of that multiplier.
Keep it predictable
- Set a daily spend cap.
- Route cheap tasks to a cheap model.
- Block runaway loops with a step budget.
- Check the spend tile daily.
A real number
Light use on a cheap model can cost cents a month. Heavy use on a frontier model, unattended, can run into double digits — which is exactly what the caps are for.