AI Chat for your Trilium share pages — drop-in widget, tool-using agent

[mrbeandev]

A small open-source project I built and wanted to share — fully complementary to the recent in-app AI chat work (#9602): this one lives on the public share page, not inside the Trilium app.

What it does

• Adds a floating "Ask the docs" chat bubble to every shared note page
• The AI uses tool calls (search_notes(query) and read_note(noteId)) to navigate your whole shared tree — only fetches relevant notes into context, so token usage stays low
  even on a 500-note KB
• Works with any OpenAI-compatible API — Gemini, OpenAI, OpenRouter, DeepSeek, Groq, local Ollama, vLLM, …
• Each visitor enters their own API key (stored in their browser localStorage) — no shared cost, no server-side key

How it works

Single JS file dropped into Trilium via ~shareJs relation. On first message it walks your shared tree using the public /share/api/notes/:id endpoint, builds an in-memory index,
and injects a Table of Contents into the system prompt. The AI then drills in via tool calls.

The widget mounts inside a Shadow DOM so its styles can't bleed into your docs page (and vice versa).

Setup

5 steps, no build step, no dependencies. Replace one ROOT_NOTE_ID constant, add ~shareJs (inheritable), done.

Full instructions, screenshots, and provider examples in the README.

Repo

https://github.com/mrbeandev/trilium-ai-agent — MIT.

Feedback / issues / PRs welcome. Especially curious if there's interest in something like this getting bundled into TriliumNext share itself eventually.