Using your own API key
MindMax can run entirely on a key you already have. There is no account to create, no weekly allowance, and no feature that stays locked — see what that covers for the whole picture and the honest limits.
This page is the mechanics.
Adding a provider
Settings → Models → LLM providers → Add provider.
The list is grouped by where the service runs — international, Chinese, and services running on your own machine. Picking one fills in its address and protocol and links you to the page where that provider issues keys.
Paste the key and save. It is encrypted using your operating system's own credential store before anything is written to disk, and it is never sent to us.
Local providers — Ollama and LM Studio — take no key at all. Start the local server first
(ollama pull a model, or switch on the server in LM Studio), then add the provider.
Choosing models
Two things happen once the address and key are in place:
- The model list is fetched from the provider itself. Not from a list we maintain — so a model released this week is there this week.
- Each tier gets a sensible default, worked out from the model names. Change any of them.
The tiers are the same ones the subscription uses:
| Tier | What it is for |
|---|---|
| Fast | Quick work — asking about an open document, tidying a paragraph |
| Standard | The default for real work: reading passes, drafting, translation |
| Deep | The passes where being right matters more than being quick |
| Vision | Reading what is on the page rather than in the text — charts, diagrams, scans |
Individual kinds of work can be pointed at a different tier in Settings → Models, so a critique pass can use a reasoning model while everyday editing stays on a cheap one.
A provider whose models do not read images cannot fill the Vision tier. Scanned-page recognition and questions about a boxed region need that tier, so if your only provider is text-only, add a second one for that tier alone.
Testing it
Test on a provider card sends one real request along the same path the app uses for ordinary work. A green result means the provider will actually answer, not that its address resolved.
Refresh models is a separate button: it only asks the provider what it offers, without sending a conversation. Use it when a provider has released something new.
When it does not work
| What you see | What it usually is |
|---|---|
| The app says no model is configured | The key is fine and the provider is reachable, but no model name has been chosen for that tier. Pick one in Settings → Models. |
| The model list stays empty | Some endpoints do not publish a list. Type the model name in by hand — once it is saved it behaves like any other. |
| The key is reported as unreadable rather than missing | Your system credential store is locked. Unlock the keychain and try again; re-entering the key will not help. |
| It worked yesterday | Keys expire and providers retire models. The error names the provider and the reason — a retired model needs a new name chosen, not a new key. |
Switching between a key and the subscription
They are not exclusive, and neither is a mode you commit to. If you are signed in, the subscription appears as one more provider in the same list, so which one does a given kind of work is decided the same way everything else is — by which provider and model that tier points at. Change a tier and the next job of that kind goes elsewhere. Nothing about your folder changes either way.