Notes library: your local notes, readable by the assistant

The Notes tab of the AI Agent page is a small local library of markdown notes with tags; the assistant searches it with read_notes and reads a note in full with read_note, and never writes to it.

Senzoukria · Documentation · Updated September 2026


Where to find it

Where
AI Agent page → "Notes" tab
Storage
localStorage key senzoukria.notes.v1, on this machine
Bounds
300 notes, 20,000 characters per body, 140-character titles, 12 tags per note
Assistant access
Read-only: read_notes (search, 8 results) and read_note (full note by id)

What it does

A note is a title, a markdown body and a list of tags. Notes are created by hand ("+ New note"), saved from a conversation with the ⤓ button of the Conversations panel (tagged "conversation", source "thread"), or filed by the assistant itself (source "assistant"). The list shows the title, the last update time, a source badge when the note was not written by hand, and the tags as #tag.

The hint under the list states the contract: "Your library. The assistant searches it before answering (read_notes) and reads a note in full (read_note) — read-only, it never writes here. Everything stays on this machine."

Search is full-text and ranked: a query word in the title scores 5, in the tags 3, and each hit in the body up to 4; ties go to the most recently updated note. The assistant's read_notes uses the same function, limited to 8 results, and receives an excerpt of about 160 characters around the first matching word.

Settings

Fields and bounds of the Notes panel
SettingDefaultWhat it changes
Title"Untitled" when emptyTrimmed to 140 characters; the graph keys wikilinks on it
Bodyempty, placeholder "Write in markdown…"Cut at 20,000 characters (MAX_NOTE_BODY)
Tags"tags, comma separated"Lower-cased, leading # removed, deduplicated, at most 12
List / Graph toggleListReplaces the editor with the notes graph; double-click a neuron to come back
Tags as nodescheckedIn graph view, each tag becomes a hub node
MAX_NOTES300Oldest notes are dropped beyond this
Deleteconfirmation "Delete the note “{name}”?"Removes the note; no undo

How to use it

  • Write your trading rules, definitions and research as notes; the assistant is instructed to call read_notes before answering a question you may already have documented.
  • Link notes with [[Title]] in the body (Obsidian convention) so the graph connects them; a title mentioned in plain text also creates a weak link.
  • Use the search field to filter the list (up to 100 results); the same query dims non-matching neurons in the graph.
  • Save a good conversation with ⤓ so it becomes searchable material instead of a thread you must scroll through.

Limits or pitfalls

  • Nothing leaves the machine unless the assistant reads a note: the content it reads then goes to the engine you selected, on your key.
  • There is no sync and no export: the store is local storage, subject to the browser quota; the 300-note cap exists to stay under it.
  • Search ignores one-letter words; a query made only of such words returns the most recent notes.
  • Reloading resets the selected note; the notes themselves persist.

This page in other languages

Frequently asked questions

Can the assistant write or edit my notes?
No. Its two tools, read_notes and read_note, are read-only. Filing is a user gesture: the ⤓ button on a conversation or the editor.
How many notes can I keep?
300. Beyond that the oldest are dropped. Each body is capped at 20,000 characters and each note at 12 tags.
How does the assistant pick which note to read?
read_notes ranks by title, tag and body matches and returns up to 8 notes with an excerpt; the model then calls read_note with the id it wants in full.

Keep reading