App shell
Tauri and Rust
A native Mac app with a Rust backend. Small, quick to open and light on memory.
Here's exactly what happens between you typing a question and Nova answering it. No step in this list leaves your computer.
Type into the Ask Nova bar, or tap a quick action like "Add task" or "Plan meeting". Plain English is fine: "move my gym session to Wednesday" works as well as a formal command.
Every message is sorted into one of four kinds: a conversation, a lookup, an action or an analysis. That decides how much of your data Nova reads and which tools it's allowed to use, so a quick chat stays quick and a big question gets the context it needs.
Before the AI model sees your question, Latis searches your notes, tasks and events itself. Each word is matched on its own and ranked by where it appears, so a match in a title counts for more than one buried in the body of a note.
Nova runs on a free, open AI model through Ollama, entirely on your computer. The answer streams in as it's written. If a model "thinks out loud" first, Latis filters that raw reasoning out, so you get a clean reply.
Every answer comes with a list of the exact notes, tasks and events Nova was given. Turn Details shows what happened in that turn: how the request was sorted, what was searched and what came back. If Nova claims it couldn't find something that the search did find, Latis flags the contradiction.
When Nova does something, it uses the same commands as the buttons in the app, with the right details passed along. Anything that deletes waits for your approval, and the approval card names the exact item first.
Open Turn Details under any answer and Latis shows you what happened behind it. If an answer looks wrong, you can see exactly why, instead of guessing.
These details are saved with the conversation, so they're still there when you come back to it later.
A native Mac app with a Rust backend. Small, quick to open and light on memory.
The screens you use every day, with types checked throughout to catch mistakes early.
Everything lives in a SQLite database on your Mac. No Latis server holds a copy.
The AI model runs on your own computer. Free to use, with no per-question charge.
Passwords and keys are kept in a Rust vault, protected with Argon2id and the macOS Keychain.
Plus a hands-on acceptance checklist on a real Mac before any stage is called done.
Latis is built from small parts with clear limits. That's what makes promises like Presentation Mode hold up: they're enforced in the structure, not just in the screen.
On your Mac. The privacy page explains every part of it.