Cost and quota
Maestro records token usage when a provider adapter returns token counters. It does not invent a number when the provider or CLI does not expose one, and it does not replace the provider’s own billing or quota page.
What can be shown
Section titled “What can be shown”The Dashboard and maestro.cmd quota can show provider health, quota windows and reset times when
the adapter can read them. Token and cost summaries can be partial, unavailable or stale. Treat an
empty or unknown reading as “not reported by this provider,” not automatically as zero.
Provider billing remains external
Section titled “Provider billing remains external”Subscriptions, API-key billing, free tiers and quota resets are controlled by Claude, OpenAI/Codex,
Google/Antigravity, GitHub, OpenCode, OpenRouter, Ollama or the configured endpoint. Maestro does
not require OPENAI_API_KEY and does not create a second OpenAI billing surface.
Reducing cost
Section titled “Reducing cost”- Set a preferred provider/model and capability order in Providers.
- Pause providers you do not want used and disable automatic fallback when strict routing is needed.
- Keep simple tasks small; Work Graphs and repeated retries can multiply provider calls.
- Use deterministic validation and checkpoints instead of starting a duplicate task after a failure.