Set up Smalt

1. Turn it on

Settings (⌘,) → Meetings → Live answers → Answered by: Jev (Recommended), or your cloud model to spot the questions too.

2. Set the cloud model

Settings → Cloud → Cloud model and keys: one model does the thinking.

1. Pick the Provider: Anthropic (Claude), OpenAI (GPT), Mistral, or Other (OpenAI-compatible) for any server that speaks the OpenAI API, with its URL.
2. Pick the Model. A fast one (Claude Haiku 5.5, GPT-6 Luna) gives the quickest answers.
3. Paste your key and click Check. It says Works, or what's wrong.

Step by step for each provider, and why Smalt uses your own key: Get your keys.

A TypeSafe key is optional: with it, Jev spots questions in a third of a second, for a few cents a meeting. Without it, the cloud model spots them.

Smalt answers questions your folders can answer, and general ones they don't cover, like "what is a design pattern?". Questions about the people present ("what do you think?", "tell us about yourself") never get an answer.

3. Choose the folders

Under Looks in, keep ~/Notes and ~/Documents/Smalt (your past meetings), or add your own. Each folder shows how many documents it holds once Smalt has read it. A document you add, save or delete there counts a couple of seconds later, during a meeting too. Answer in picks the language of the answers: the question's, by default.

4. Try it

Start notes and ask something your files answer. The answer shows up under the question a few seconds later. Or hold ⌥⇧ and ask it yourself.

See what it does

Turn on Judgment log and click Open Log. Each line shows what was judged, what was found and how long it took:

16:24:37  #2  0.88  Et pourquoi Tom Sawyer, elle avait quel âge ?
              ⌕ 23 - The Adventures of Tom Sawyer.txt
              Card · GPT-6 Luna
              spot 0.2 s · search 0.1 s · answer 1.9 s · total 2.2 s

Each meeting gets its own log; the menu at the top of the window goes back to earlier ones, and the last 30 are kept. The log holds meeting text, so it's off by default.