Get started

Choose your AI

Minne has no model of its own. Sign in with a subscription you already pay for, bring an API key, or point it at a model on your own machine.

Minne is the memory and the machinery; the thinking is done by a model you choose. There is no Minne account and no Minne billing, because there is no Minne server for a model to live on.

The choice lives in Settings → Account ("Which AI Minne thinks with"), and it is the first thing first-run asks you. Four cards:

One model serves everything — chat, the Minne key, and the sync pass that distils captures into wiki pages. That is one function in the code rather than a convention, so there is no way for the three to drift apart.

#Signing in with a subscription

Pick the card and press Sign In. Minne opens your browser at the provider's own authorization page; you approve there, and the browser hands a token back to a listener on 127.0.0.1. Minne never sees your password.

If the browser hand-back does not complete — some setups block the loopback listener — the sign-in also accepts the authorization code pasted back into the app. Whichever finishes first wins.

#Using an API key

Pick API key, choose Anthropic or OpenAI in the "Key from:" popup, paste the key, and press Use This Account. Nothing goes over the network during sign-in — the key is written straight to disk.

It lands in ~/Library/Application Support/Minne/auth.json, a file created 0600 in a 0700 directory: readable by your user account and nobody else. Billing is the provider's, per token.

#Running a local model

Pick Local (Ollama), type the server address, press Use This Server. There is no sign-in step, because there is nobody to sign in to.

The default address is http://localhost:11434/v1 — Ollama's OpenAI-compatible endpoint. Any OpenAI-compatible server works: vLLM, LM Studio, whatever you run. Minne talks /chat/completions to it.

Two things worth knowing before you commit to local:

• The model must be able to call tools. Every Minne feature runs an agent loop with the memory tools attached; a model that cannot make tool calls cannot read or write the wiki, and nothing in the app will warn you about it.
• The default local model is llama3.1, and there is no picker for it. The card's only field is the server address. To use a different local model, edit ollama.model in config.json — see CLI and environment.

In exchange: nothing leaves the Mac. Not the captures, not the drafts, not the questions.

#Choosing a model

The provider card has a Model: popup listing that provider's catalogue. It works before you sign in — the catalogue is static — and changing it while the provider is already in use applies immediately, with no re-sign-in.

Left alone, each provider uses a sensible mid-tier default: claude-sonnet-5 for Anthropic, gpt-5.5 for OpenAI. Mid-tier on purpose: the same model does chat, drafting and the background distillation, and the distillation runs whether you are watching or not.

#Switching, and signing out

Switching providers is picking a different card. Signing out clears the stored credential for the current provider.

Revoking the grant on the provider's side (in your Anthropic or OpenAI account settings) is a separate act, and one Minne cannot do for you.

#What each choice sends where

Every path, in detail, is on What leaves your Mac. In short: model requests go to api.anthropic.com, chatgpt.com/backend-api, api.openai.com, or the local address you set — and to nowhere else.