Trust and cost
What it costs
Wachiman is not a service you subscribe to; model calls use your provider account, and the engine budgets each review.
Wachiman has no plan, seat or invoice because there is no service. You pay your model provider for the calls the review makes, plus any GitHub Actions charges that apply to your account. The amount depends on the size of the pull request, the model and your provider’s prices.
What the engine limits
Section titled “What the engine limits”The engine groups files into batches ordered by path-based risk and budgets each review. Three details change what that budget means:
- The batch limit applies per call, not per batch. A batch whose first call leaves files unaccounted for may get one retry. A retry is a paid second call, and each call is capped separately, so one batch can cost more than the batch limit.
- Enforcement depends on the provider. For Claude Code, the engine passes the dollar cap to Claude Code, which enforces it as a soft limit. The Codex adapter, which is CLI-only, has no dollar cap at all; only a time limit bounds a call.
- Spending stops, and coverage says so. Once the remaining review budget is nearly spent, later batches do not run. A review that ended partway says so on its first line.
Do not treat the configured numbers as an absolute billing ceiling. They are limits on engine behaviour, not a guarantee from your provider.
The budget limits one run, not a month and not the life of a pull request. If your caller lists
synchronize, each push to an open pull request starts its own review, and each of those reviews
can spend up to the review budget.
Seeing what you spent
Section titled “Seeing what you spent”Completed batch reviews show the model and the reported cost in their footer, and partial reviews identify the missing coverage. Early-exit notices, such as “Already reviewed” or a review that ran out of budget before any batch finished, have different output and no cost footer, even though a budget-exhausted run may already have spent money.
For spend across repositories and over time, use your provider’s usage console. Which breakdowns and spending controls it offers depends on your provider and account; check how they are enforced before relying on them.