Ito 糸

Thread. The invisible line between things that belong together.

---

What Ito Is

Ito is an Obsidian plugin that surfaces semantic connections across your vault — across notes, images, PDFs, audio recordings, and video — without you needing to tag, organise, or link anything in advance.

It uses Google's Gemini multimodal embedding model to map every file in your vault into a single unified semantic space. A voice memo, a scanned diagram, a markdown note, and a PDF paper can all be compared against each other. When you open a note, Ito quietly finds what belongs near it.

---

The Ideology

Your knowledge base should only contain what you actually know.

There is a growing category of AI tools that write into your knowledge base for you. They summarise articles, generate notes, compile wikis, synthesise connections. The result is a vault full of text — but whose understanding does it represent?

The more an AI writes into your knowledge base, the more it pollutes it. You end up with a collection of LLM outputs that look like knowledge but carry none of the cognitive weight of having actually learned something. You cannot teach it. You cannot defend it. You do not know where half of it came from. That is not a knowledge base. That is a cache of hallucinations dressed up as your understanding.

Real knowledge requires the effort of formation. Writing a note forces comprehension. You cannot write clearly about something you do not understand. The friction is not a bug — it is how knowledge becomes yours.

Ito's position is the opposite of that trend.

Every note Ito surfaces was written by you. Every connection it suggests is between your words and your ideas. The moment of recognition when two of your own notes belong together is real learning — because you already understood both things. Ito just reminded you they were connected.

Ito does not write. It does not summarise. It does not generate. It resurfaces.

The model finds candidates. You confirm connections.

---

Ito as a Knowledge Base & Retrieval Tool

Ito is not just a sidebar panel. It is the foundation of a two-part system for navigating your vault from anywhere — including inside an AI conversation.

The setup

1. Ito plugin (this) — indexes your vault, keeps a live vault-graph.json in your vault root
2. ito-cli — reads that file from any terminal

# Install the CLI
curl -O https://raw.githubusercontent.com/heetvadiya/ito-cli/main/ito.py

# Set your vault once
python3 ito.py --config-vault ~/obsidian/MyVault

# Add alias so it works everywhere
echo 'alias ito="python3 /path/to/ito.py"' >> ~/.bashrc
source ~/.bashrc

# Search your vault from anywhere
ito "developer experience"
ito "what do I know about embeddings"

Use with Claude or any AI

Add this to your Claude system prompt or CLAUDE.md in any project:

When asked "what do I know about X" or "find my notes on X",
run: python3 /path/to/ito.py "X"
Answer from the output.

Now asking "what do I know about knowledge management?" makes Claude search your actual vault and answer from your own notes — not its training data.

ito outputs structured markdown — headers, paths, tags, links, and full note content. LLM-friendly by design. No parsing, no preprocessing. The model reads it directly.

---

Dark mode: Light mode: Settings Panel:

What Ito Does Not Do

• It does not generate text.
• It does not summarise your notes.
• It does not write links into your vault without your explicit action.
• It does not send your files to any server except the Gemini API embedding call, using your own API key against your own quota.
• It does not modify any file in your vault except appending to a ## Related section when you explicitly click Add link.

Your vault stays exactly as you left it.

---

Multimodal

Gemini Embedding maps text, images, audio, video, and PDFs into the same vector space. This means:

• A sketch you photographed can surface the note it belongs to.
• A voice memo from a walk can connect to the essay you wrote a week later.
• A PDF paper can relate to your handwritten notes on the same topic.
• An image embedded in one note will be indexed as part of that note — so results always surface the note itself, not the raw file.

All of this happens without transcription, without description generation, without any LLM pass over your content. The embedding model reads the raw signal directly.

---

Features

Semantic Panel

Open with Ctrl+Shift+I. A sidebar panel showing the top related files for whatever note you are currently reading. Results include a matched content snippet so you can see why a connection was suggested before deciding to act on it.

The similarity threshold slider re-filters results in real time without any API call — the panel works entirely from cached embeddings after the initial index.

Find Similar to Selection

Highlight any text in a note and run Ito: Find notes similar to selection from the command palette. Embeds the selection on demand and opens a results modal.

Vault Indexer

Runs silently in the background. On first load, reconciles the entire vault against the index. After that, re-indexes files when you save them. New files are picked up automatically.

Progress and any errors surface as Obsidian notices. Nothing happens invisibly.

Add Link

One button in the panel adds [[filename]] to a ## Related section at the bottom of your active note. That is the only moment Ito writes anything to a vault file — and only when you ask it to.

Vault Graph Export

Ito automatically maintains a vault-graph.json file in your vault root. It contains every note's title, links, backlinks, tags, and a content summary — the same data Obsidian uses to render the graph view, exported as clean JSON.

The file refreshes automatically whenever:

• Obsidian opens
• Any note's links change
• A file finishes indexing

You can also trigger it manually: Ito: Export vault graph in the command palette.

This file is the bridge to ito-cli — a companion command-line tool that lets you (or any AI) navigate your vault without opening Obsidian.

---

Setup

From the Community Plugin Directory (pending review)

1. Open Obsidian → Settings → Community plugins → Browse
2. Search for Ito and install
3. Enable the plugin
4. Go to Settings → Ito, paste your Gemini API key
5. Click Test connection to verify
6. Ito will begin indexing your vault automatically

---

Manual Install (available now)

Requirements: Node.js 18+, Git

# 1. Clone the repo
git clone https://github.com/heetvadiya/ito.git
cd ito

# 2. Install dependencies and build
npm install
npm run build

This produces main.js in the project root.

# 3. Copy the three plugin files into your vault
cp main.js manifest.json styles.css /path/to/your/vault/.obsidian/plugins/ito/

Replace /path/to/your/vault with the actual path to your Obsidian vault. Create the ito folder if it doesn't exist.

4. Open Obsidian → Settings → Community plugins → disable Safe mode if prompted → enable Ito
5. Go to Settings → Ito, paste your Gemini API key
6. Click Test connection to verify
7. Ito will begin indexing your vault automatically

Model used: gemini-embedding-2-preview
Supported files: .md, .txt, .png, .jpg, .jpeg, .webp, .pdf, .mp3, .wav, .m4a, .ogg, .flac, .mp4, .mov, .webm

---

Settings

Setting	Default	Description
Gemini API key	—	Required. Never stored outside your vault's local data.
Indexed folders	(entire vault)	Comma-separated paths to scope indexing.
Embedding dimension	768	768 / 1536 / 3072. Changing requires full re-index.
Similarity threshold	75%	Minimum score for a result to appear.
Max results	8	Between 3 and 20.
Auto-index on save	On	Re-indexes a file when you save it.
Index audio files	On	MP3, WAV, M4A, OGG, FLAC.
Index video files	Off	MP4, MOV, WEBM. Opt-in — large files consume significant quota.
Max file size	50 MB	Files larger than this are skipped silently.

---

Privacy

Ito sends file content to the Gemini API for embedding. This is the only external call the plugin makes. Everything else — the vector index, similarity computation, results ranking — runs locally inside Obsidian.

The index lives at .obsidian/plugins/ito/ito.db. It travels with your vault and belongs entirely to you.

---

The Name

糸 — Ito — is Japanese for thread.

A thread is the connection before it becomes a knot. It is potential, not structure. Ito finds the threads. You decide which ones to tie.

---

Built with intentionality. The connections are yours.