Skip to main content
AI models process text as tokens. Otto shows token details when the provider reports them, but you’re billed for verified usage, not a token-rate formula.
  • Input tokens are the context sent to the model: the diff and surrounding code, the ticket, instructions, conversation history, and tool results.
  • Output tokens are what the model writes: review findings, plans, code, and replies.
Some models also report cache or reasoning tokens.

​
What drives usage

​
Keeping usage down

  • Match depth to risk. Run standard by default and raise only critical paths. See Review depth.
  • Ignore what doesn’t need review. Snapshots, fixtures, and generated code. See Ignoring files.
  • Skip PRs that don’t need review. Bot updates and release merges. See Skipping reviews.
  • Keep instructions focused. Include the conventions that matter; leave out unrelated documents.
  • Split large work into smaller PRs and tickets. They’re cheaper and easier to review.