Costs and budgets
Commandr counts every request, so you always know what a project has cost — and can cap it.
In short
- The project’s cost is always in the chat header.
- Set a budget: you’re warned at 80%, and automatic work stops at 100%.
- Local models are free.
The cost meter
Every model request is counted: input tokens (including cached ones), output tokens and cost, per reply, per agent and per project. The project’s total is in the chat header; open it for the breakdown by agent and model. Each reply shows its cost next to its time.
Where the numbers come from
Costs come from the usage each provider reports and a price list built into the app, including cache discounts and DeepSeek’s peak-hour prices. OpenRouter reports its own cost. Local models are free, models without a known price show tokens only, and providers that don’t report usage are estimated.
Treat the numbers as close estimates: your provider’s bill is the final word.
Set a budget
Open the cost meter in the chat header and set a budget for the project.
- At 80%, you get a warning.
- At 100%, the pipeline, task auto-run and fix rounds stop. Agents you run yourself still work.
To carry on, raise the budget in the cost meter.
What counts
- Every agent reply, including fix rounds and visual checks
- Starter suggestions in empty chats
- Stopped replies, as an estimate — the provider still bills them