Usage and costs
Maestrly App is free, open-source software under the MIT License, currently in 0.x releases. There is no Maestrly account, paid tier, or hosted backend. AI inference and connected services use accounts you own and may charge you under their own pricing.
Who bills what
| Service | Account and billing |
|---|---|
| Maestrly | Free access; no paid tier or paywall currently. |
| AI provider | Your own account, subscription, credits, or API billing. |
| MCP or connected service | Your own third-party account and that service's plan, when applicable. |
| Bot server (optional) | Self-hosted on this computer's Docker or on a VPS you rent. You pay for that machine: hosting, compute, storage, and traffic. Bots, and chats and workspaces you run there, use the model accounts you add to their environment. |
| Voluntary contribution | Optional support for Maestrly; it does not buy usage, features, credits, or priority support. |
Maestrly is not affiliated with AI providers and does not set their prices, quotas, or model availability.
See usage in Maestrly
- Subscription usage: when a Claude, ChatGPT, or Google AI subscription is connected, Usage at the bottom of the sidebar shows each plan's current limits, such as the 5-hour window and weekly windows, with the percentage used and when each resets. Settings → Maestrly Chat → Accounts shows the same Usage limits per account.
- Context meter: each conversation shows how much of the model's context window is in use. Hover over it to compare the last measured value with the estimate for the next request, marked with
~, and to see the accumulated estimated cost. - Settings → Usage & cost: tokens and estimated cost per model on this device, for Today, 7, 30, or 90 days, or a custom range within the last 90 days. Models without a catalog price are left out of the total.
- Subagents: each subagent session shows its own estimated cost.
- Bots: a bot's composer shows its context use and estimated cost.
Costs are estimates based on what the runtime records and public models.dev pricing. On a flat subscription they do not necessarily match what you are charged. The provider's dashboard, invoices, and quota pages are the source of truth for billable requests and remaining limits.
For each connected service, know:
- which account or organization is active;
- whether billing is subscription, credit, request, token, or tool based;
- the reset period and rate limits;
- who can change budgets or credentials.
Control context
- Compact a long conversation, or type
/compact, to summarize earlier history and free context. Maestrly suggests it when the context fills up. - Click the context meter to set a Context limit for the current model on that account, below its real window, or No limit to remove it. Since 0.23.0, it works with API providers and with the Codex, Claude, GitHub Copilot, Grok, Cursor, and Google AI subscriptions (Cursor and Google AI compact between turns; ChatGPT Web has no limit).
- Prepare context, under Background tasks in Settings → Maestrly Chat → Models & agents, prepares summaries with a separate model you choose. It starts off, and each preparation uses that model's quota.
- Bots (Experimental) can cap their window in Bot settings → Model → Compaction → Maximum context window (thousand tokens), or through their environment's default. The bot compacts at 90% of the cap; restart the environment to apply a change.
Control cost
- Choose a model proportional to the task. Fast mode answers faster with higher usage, and the highest reasoning efforts are slower and more expensive.
- Subagents, Maestro, code and plan reviews, parallel conversations, and a bot's parallel tasks or Delegate mode each make their own model calls. Limit them, and limit retries.
- Some features use quota in the background: context preparation, saving memories from conversations, image interpretation for models without vision, and bot compaction. Image generation for models other than Codex consumes your connected ChatGPT or Google AI subscription (Google AI images report no token usage). Each bot routine run is a full model turn, and since 0.22.0 a bot's runs and conversations with other bots can run at the same time as its conversation with you.
- Voice dictation through OpenAI or Groq is billed to that provider account, about US$0.006 and US$0.0007 per minute; This computer (Whisper) costs nothing.
- Turn off Maestrly tool groups and MCP servers a conversation does not need; their descriptions take up context. Compact command output trims verbose shell output.
- With several Claude or ChatGPT accounts, Automatic rotation retries on the next account when one hits its usage limit.
- Test agents, skills, and tools manually before automation, and stop or pause workflows that repeatedly fail.
- Set budgets and alerts in provider consoles when available.
A failed, retried, or canceled request may still be billable under the provider's rules.
Handle limits and availability
If a request is rejected, distinguish authentication, rate, quota, billing, regional, and model-availability errors before retrying. Switching models may change quality, capabilities, context size, and price. Provider status and documentation are the source of truth.
Voluntary contributions
You may support Maestrly with an optional one-time or monthly contribution. Contributions do not unlock product access or increase provider limits, and they are never converted automatically into a product subscription. See the Contribution & Refund Policy for cancellation and refund details.
Next: Chat, models, and context.