Desktop and phone

Using AI on a note

Insights, Polish and Chat, applied to the note you are looking at.

The AI Magic button in a note header opens a panel with three tabs. All three run against the model you configured in Connect an AI model.

Insights, Polish and Chat, docked beside the note.

Insights

Insights reads the note and returns a summary, action items, themes, and anything it thinks is missing. Pick a template first: General Insights suits most notes, and there are purpose-built ones for journals, health logs, finances and meetings.

Insights running against a local model. The whole exchange stays on this machine.
The result, with Copy and Append to note underneath.
Nothing is written to your note without asking
Insights output sits in the panel until you press Append to note. Close the panel and it goes away.

Deep Insights

For a note that matters, Deep Insights runs the same template in three passes instead of one: it first lists every figure, date, name and decision in the note as a ledger, then searches your library for related notes (reading at most five) and names the connections and contradictions, then writes the report over all of it with the rule that nothing in the ledger is dropped. It is opt-in for each run, and the button says what it will cost before you press it.

Polish

Polish rewrites the note in one of six registers: Standard, Verbose, Succinct, Gentle, Formal or Casual. The rewrite appears in the panel next to an Apply to note button, so you read it before it replaces anything.

Six registers to choose from.
Read it first. Apply to note is a separate, deliberate step.

Chat about this note

The Chat tab talks to the current note. Choose an assistant personality (Balanced, Concise, Encouraging Coach, Socratic Guide, Casual Friend, Analytical Strategist, or one of two character voices: Grimblek the Destroyer and the ADHD Addled Psych Professor), then ask.

Every answer carries a What was sent? line showing the exact token counts and context window.
You can always see what was sent
The usage line under an answer, and the context inspector behind it, show precisely what went to the model. Nothing is hidden.

On the phone

Pick a template.
Running, with Stop always available.
The result, with Regenerate, Copy and Save.
AI Magic on Android, running against a model on the local network.

Templates

Every built-in template can be edited, and you can add your own, under Settings ▸ AI Templates. A SILO can also have its own default template, so your journal gets a journal prompt and your work notes get a work one.

Insights, polish and chat templates, all editable.

Thinking mode

With a reasoning model, Thinking mode shows the model working before it answers, in a collapsible panel. It is slower and usually better. The toggle is in AI settings and in the AI Magic header.