Skip to main content
bolt stats aggregates token counts and cost across the local session database. Because every tool call is recorded (see Local Database), the numbers are exact for the LLM calls Bolt drove.

Basic usage

What it shows

  • Total input, output, and cache tokens.
  • Estimated cost per provider (using provider pricing tables at the time of the run).
  • Per-model, per-agent, and per-day breakdowns depending on flags.

Reducing cost

  • Use --variant minimal or --variant low on reasoning-heavy models. See Providers & Models.
  • Prefer the ask or plan agent for exploration; they’re read-only and avoid tool-heavy loops.
  • Switch to a smaller model with bolt models for tasks that don’t need frontier capability.