Notes and AI

Two different things work on notes: background automation that quietly keeps the base in order, and the agent, which acts when you ask. The distinction matters — they have different rights.

What happens in the background

After a note is created or substantially edited it enters a background queue. There it gets:

a description — a short digest, shown in lists and useful for finding the note by eye;
tags — from the content, so you don’t place them by hand;
other card metadata.

The same pass grows concepts and links: people, projects, companies and topics are extracted from the text, and edges appear between them and the note. That’s why the agent can answer “what’s happening on project X” with more than exact word matches.

In parallel the note is indexed for search — full-text and vector. Indexing is local.

What the agent can do

On request, in chat, the agent can:

find what you need by meaning rather than by exact word;
read a specific note and answer from it, citing the source;
create a note — for example, writing up the outcome of a conversation;
modify an existing one;
delete one.

Creating, modifying and deleting are changes, so the agent asks for confirmation. Reading and searching happen without questions.

Notes from other modules

From a recording or meeting — the transcript and summary are saved as a separate note, marked as automatically created.
From an email — a message becomes a note, and the link to the original is kept.
From a plugin — an installed plugin with write access to notes creates them through the same path as the agent, and leaves a trace of its origin.

Origin is always visible: a note knows whether it came from mail, a meeting, the agent or your hands.

Privacy

Before anything goes to the cloud, the text passes through sensitive-data masking — card, account and password numbers are replaced with placeholders. If cloud features aren’t connected, notes never leave the device: search, indexing and the local model all run on it.