Lorelai hears, thinks and speaks entirely on-device — with memory that persists, profiles per app, and hands into your tools. Nothing you say ever leaves the machine.
Move my 3 pm with Ola to tomorrow and tell her I’m running late on the deck.
Done — Ola is now Thursday 3 pm. The note to her is open in Mail, ready for you to send.
No app to open, no tab to switch to. Lorelai sits in the menu bar and drops in as a ribbon over whatever you’re doing.
Hold your hotkey anywhere — you choose the keys. The ribbon morphs open and starts listening; release to send. Speech is transcribed on-device, in real time.
Your hotkey · any appA language model running on your Apple Silicon reasons over your request, what’s on screen, and what Lorelai remembers about you. Latency is bounded by your chip, not by a queue.
On-device · Apple SiliconCalendar, Mail, Contacts, files, todo apps — Lorelai calls the tool, shows a pill for each action, and answers back in a natural on-device voice.
Tool pills · Local voiceEvery model Lorelai uses — speech recognition, language model, voice, embeddings — ships inside the app and runs on your Apple Silicon. There is no Lorelai server that hears you. Unplug the network and nothing changes.
Memory works the same way: facts about you and past conversations live in a database on your own disk, which you can inspect, edit and wipe. It never leaves the machine — and if you turn on sync between your own Macs, it is encrypted end-to-end before it goes anywhere.
Native macOS frameworks come built in. Cloud services connect through OAuth straight from your Mac — tokens stay in your Keychain.
200 lifetime licenses at a launch-only price for the people who get in early and tell us what’s wrong. Waitlist members get first access, in order of signup.
Yes, once the models have downloaded the first time. Transcription, reasoning, memory and voice all run on your Mac. Only the optional cloud integrations (Todoist, web search) need a connection — and they talk to those services directly, never through us.
Any Apple Silicon Mac (M1 or newer) on macOS 14 or later. Lorelai picks a model to fit the machine: 8 GB of memory runs the small, fast one; 16 GB runs the standard model; 32 GB and 64 GB unlock the larger ones. More memory buys better answers, not more features.
Siri can’t remember across sessions or run per-app behaviors; ChatGPT voice sends everything to the cloud. Lorelai keeps persistent, editable memory and acts inside your Mac apps, without a server in the loop.
Yes — dictation, translate, conversation with session memory, one integration and two per-app profiles. No time limit. Pro Lifetime removes limits and adds persistent memory.
No. Deep integrations like AppleScript, Health and OAuth tokens don’t fit App Store rules. Lorelai is sold directly, notarized by Apple.
Private beta for the first waitlist cohort this autumn; public launch to follow. You’ll get exactly two emails from us.