Local AI on your Mac, done right. Private by architecture.
Routines runs a local-first AI assistant for normal people. Your notes, meetings, and memory stay on your Mac as plain markdown and SQLite. Run a local LLM, bring your own key, or use the AI included with your account. Your notes, meetings and memory stay on disk in every case; only the model call travels, and the page below says exactly where.
Most "private" cloud apps just promise not to look. Routines keeps your notes, meetings, and memory on your Mac as plain markdown and SQLite, so privacy is the default shape of the app, not a setting you hope someone honored.
02
Run a local LLM, your own key, or ours
Point Routines at a local model, bring an Anthropic API key, sign in with your ChatGPT or Claude subscription, or use the AI included with your account. A local model never leaves the Mac. Your own key lives in the macOS Keychain and the request goes out from your machine, never through us. Routines AI passes the request through our server to Anthropic and stores no part of it.
03
Memory you can read in Finder
Your assistant remembers context across days because that memory is markdown files on disk. Open them, edit them, back them up, delete them. No hidden vector store you cannot see.
04
Works whether you are online or not
Reading your routines, outputs, meetings, and memory hits your local database and works with the network off. Only the model call leaves your machine, and only to the provider you chose.
Cloud AI vs local AI
Why local beats a cloud toggle.
Typical cloud AIRoutines
Where your data livesOn a vendor server you cannot inspectOn your Mac, as markdown + SQLite
Who holds the API keyThe vendor, bundled into a subscriptionYou, in your macOS Keychain
Privacy modelA promise and a settings toggleLocal by architecture
Offline behaviorMostly unusableLocal reads work anytime
Local AI means the app and your data run on your own computer instead of a company server. Routines stores everything on your Mac as markdown and SQLite, and only sends a request to a model provider when you actually ask the AI to do something.
Can I run a local LLM with no internet at all? +
Routines can read your routines, meetings, memory, and outputs fully offline because they live in a local database. The text-generation step needs a model, and that step needs a connection: point it at a local model you run yourself, or, when you are online, let Routines AI handle it or use a cloud provider on your own key.
Do you ever see my data or my API key? +
We never see your key: it lives in the macOS Keychain and is never sent to us. Your notes and memory stay on your Mac. The prompt is the one thing that travels, and there are two shapes it can take. On your own key or a local model, the request goes from your machine to the provider and there is no Routines server in the middle. On Routines AI, the request passes through our server to Anthropic and we store nothing of it, only the tokens and cost we bill.
How is this different from a cloud AI with a privacy mode? +
A privacy toggle asks you to trust a promise about data the vendor is already holding. Routines holds none of it: your notes, memory and history are files on your disk, and the only thing that ever moves is a single prompt, either through a server that stores nothing or straight to the provider on your own key.
Keep your AI, and your data, on your Mac.
Local-first by design. AI included, or run a local model or your own key. Free for 7 days.