Settings reference
Settings opens from the app menu or with ⌘, (Ctrl+, on Windows). Pages that depend on a
knowledge base — knowledge base, memory, translation memory — show the one currently open.
General
- Appearance — dark, light, or follow the system.
- Language — the interface language. This does not decide what language the AI answers in; that follows the language you write to it in.
- Agent memory — switch automatic memory on or off. Off means nothing new is captured and nothing is injected; what is already stored stays until you delete it.
- Keep running in the tray — when the last window closes, MindMax stays running so the global voice shortcut still works.
The page also shows the installed version and Check for updates.
Account
Your plan, your weekly credits, and the devices signed in. An account is only needed for the subscription — running on your own API key needs none. Credits are granted weekly; unused credit carries over, up to two weeks' worth.
Usage
Two tabs that measure different things and are never added together:
- Subscription credits — what MindMax metered for calls through its own models, with the weekly balance, a daily trend, and breakdowns by tier and by model.
- Model usage — raw token counts recorded on this machine for every model call, including calls on your own API keys, by provider and by task type. Exportable as CSV.
Reading, highlighting, searching and exporting your own documents spend nothing.
Models
Two tabs.
Model selection decides which model each kind of work uses. Work is grouped into tiers — Fast, Standard and Deep for text, plus dedicated tiers such as Vision for reading images — and each feature (chat, reading passes, translation, canvas, recognition and so on) is assigned to one. Leave them on Auto unless you want a particular model somewhere. On the subscription every plan, the free one included, can use Fast, Standard and Vision. Deep starts on Starter, in a limited amount, and is fully included from Pro; on the free plan, Deep work runs on the best tier your plan includes, and the page tells you so. Deep-tier work uses up credits faster. With your own key, you choose what sits in each.
LLM providers is where you add your own keys and local models. Test sends one real request along the path ordinary work uses; Refresh models only re-fetches the provider's model list. Keys are encrypted with your operating system's credential store before they are written to disk. See using your own API key.
Editor
- Display — the "Type / for commands" hint on empty lines, and an editable file name at the top of a note.
- AI autocomplete — a grey continuation suggested when you pause while typing; Tab accepts,
Esc dismisses. Off by default. Even when off,
⌘/(Ctrl+/) asks for one suggestion. - Selection toolbar — the floating toolbar with formatting and AI actions that appears when you select text.
- Revision author — the name on tracked changes when AI revisions are written back to a Word file, and on comments in an exported reviewed copy. Left empty, it uses your account name or your computer's user name.
- Translation languages — add target languages to the list offered by Translate in the editor's AI menu, beyond the built-in ones.
Knowledge base
What the AI can and cannot read in the open folder:
- Coverage — how many of your files the AI can search, how many are still being read in, and which need you: images not yet read, scans without a text layer, files that could not be read (with Retry).
- Files and folders hidden from AI — still in the file tree and in search, skipped by AI retrieval.
- Folders that aren't part of this knowledge base — excluded entirely, from the tree, search and AI alike.
- How each feature uses the knowledge base — the default for chat, writing, the canvas, the editor's AI and translation, spelled out.
- What MindMax added to this folder — the three things it writes, all deletable.
See choosing what the AI can read.
Memory
What the assistant has remembered, as plain-text entries you can read, edit and delete — split into memories for this knowledge base and global ones that apply across all of them. Each entry can be limited to particular tasks. See history and memory.
Translation memory
Term translations settled in this knowledge base, per language pair. The next document in the same pair reuses them first. An entry you edit becomes confirmed and is never overwritten automatically.
Hotkeys
Every command with its shortcut. Filter, change, disable or reset any of them. The Voice note
(global) shortcut — ⌥⇧Space on macOS, Alt+Shift+Space on Windows by default — works even
when MindMax is in the background. Shortcuts handled inside the editor are listed for reference.
Hold ⌘ (Ctrl on Windows) for about a second anywhere in the app to see the shortcuts available
right now; release to close. When you do the same thing with the mouse a few times, a small hint
next to the button shows its shortcut — at most two a day, and never again once you have used that
shortcut. Both can be turned off under Learn shortcuts.
Skills
Built-in and installed skills for chat, the canvas and the editor: copy a built-in to edit it, hide the ones you never use, install examples from the starter library, import a skill file or export one to share. See using and writing skills.