dotstell

What is Dotstell?

Dotstell is an open source personal knowledge graph — a single place to write notes, track people, manage tasks, save bookmarks, and connect all of them together through a visual graph. With a built-in AI layer that understands your entire knowledge base.

Most productivity tools are built around capture. Dotstell is built around connection — and now, conversation.

Capture your ideas, meeting notes, and research. Connect them to people, tasks, and bookmarks — building a knowledge graph that grows with you. Ask an AI that answers from your own notes, not the internet — chat, get person intelligence, generate writing, and discover what you already know.

Everything you write, everyone you know, everything you do — all linked in one living knowledge graph. Now with AI that understands all of it.

The question it answers that other tools don't:

"What do I know about this person, this project, this decision — right now, in full context?"

Support the project

Dotstell is free, self-hostable, and built in the open. If it saves you time or you believe in what we're building:

• Star the repo — the single most useful thing you can do. It helps others find Dotstell and tells us the work is worthwhile.
• Open an issue — bug reports and feature requests shape what gets built next.
• Contribute code — all skill levels welcome; see the contributing guide.
• Share it — tell someone who needs a better way to organise their work.

Every star, issue, and PR genuinely matters for an early-stage OSS project.

Features

Capture

Connect

Ask

Yours to keep

Screenshots

Live previews of the app are available on dotstell.com — the landing page includes an interactive knowledge graph, dashboard mockup, and feature walkthroughs.

Try the live app at dotstell.app or download the desktop build for Windows, macOS and Linux.

Getting Started

Prerequisites

• Node.js 20.9+
• pnpm — npm install -g pnpm

  We use pnpm instead of npm/yarn because it is significantly faster (hard-linked global store, no re-downloading the same package twice), produces a deterministic pnpm-lock.yaml, and is required by the Tauri CLI toolchain used for the desktop build. The lockfile is committed — always use pnpm install, not npm install, to keep the lockfile consistent.

• A free Supabase account

1. Clone

git clone https://github.com/dotstell/dotstell.gitcd dotstell

2. Install dependencies

pnpm install

3. Set up Supabase

1. Create a new project at supabase.com

2. Go to SQL Editor and run supabase/schema.sql to create all tables, policies, and triggers.

   That one file is all a new project needs. For an existing database, apply any migrations you have not run yet, in order:

supabase/migrations/├── add_notebooks_table.sql├── add_notes_soft_delete.sql├── 002_bookmarks_enhancements.sql├── 003_bookmark_visits.sql├── 004_notes_parent.sql├── 005_notes_sort_pin.sql├── 006_notes_color.sql├── 007_notebooks_constraints.sql├── 008_ai_embeddings.sql          ← vector(768) columns on notes + bookmarks (AI semantic search)├── 009_ai_match_functions.sql     ← pgvector cosine similarity RPC functions├── 010_tasks_embedding.sql        ← vector(768) column on tasks + match_tasks() (tasks in AI Chat RAG)├── 011_fix_function_search_paths.sql├── 012_mcp.sql                    ← MCP tokens and the read-only functions behind the MCP server└── 013_data_api_grants.sql        ← explicit table grants (needed on new Supabase projects from 30 Oct 2026; safe to run on any)

3. Copy your project URL and anon key from Settings → API

4. Configure environment

cp .env.local.example .env.local

Edit .env.local:

NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.coNEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key

Behind a corporate SSL proxy? Add NODE_TLS_REJECT_UNAUTHORIZED=0 to your .env.local. Do not set this in production.

5. Run

pnpm dev

Open http://localhost:3000, create an account and start connecting your knowledge.

AI Features

Dotstell ships a complete AI layer. All features work with five provider options:

AI features:

• Meeting assistant — paste or upload a Zoom, Teams or Meet transcript (or record the meeting) and get a summary, decisions and action items; turn action items into linked tasks
• Voice dictation — Fast (your browser, with spoken punctuation and automatic capitals, English) or High accuracy (your own OpenAI, Groq or Gemini speech key, with punctuation and capitals from the model); both work on phones
• Connect AI tools (MCP server) — let Codex, Claude Code, Cursor and Claude Desktop read your notes, bookmarks, tasks, people and meetings with a revocable, read-only token (see docs/MCP.md)
• AI Writing Assistant — draft notes from scratch using 8 starter templates (Outline, Meeting notes, Daily log, Research note, OoO email, Proposal, Status update, Email draft) or a custom prompt; improve existing notes with one-click actions (Improve English, Make formal, Make concise, Expand, Full rewrite)
• AI Chat — slide-out panel with RAG (semantic search grounds answers in your notes, bookmarks, and tasks), "This note" and "All knowledge" modes, People tab for person intelligence
• Inline Assist — select text → rewrite / expand / shorten / fix grammar / outline / checklist / explain
• Smart title & auto-tags — suggest a title and up to 5 relevant tags as you write
• Note & bookmark summaries — one-click summary in bullet or paragraph form
• AI Knowledge Digest — dashboard panel summarising your latest notes activity
• Semantic Related Notes — sidebar panel showing notes similar to the one you're reading
• Person Intelligence — searches all your notes and bookmarks for mentions of a person and generates a structured brief
• Graph intelligence — finds missing links, note clusters, and bridgeable gaps in your knowledge graph

API keys are stored only in your browser's localStorage. dotstell's servers never store or log them: a request that goes through a dotstell API route carries the key for that one request and passes it to your chosen AI provider over TLS. Speech-to-text (recording and High accuracy dictation) goes straight from your browser to the provider.

Using Ollama with the live app — Local AI Agent

The hosted app at dotstell.app runs over https://. Modern browsers block https:// pages from fetching http://localhost directly (Private Network Access spec). To use Ollama with the live app, run the Dotstell Local AI Agent — a tiny zero-dependency Node.js proxy that adds the required browser security headers.

Follow these steps in order:

Step 1 — Install and start Ollama

Download Ollama from ollama.com if you haven't already, then make sure it is running:

• Windows: Ollama usually auto-starts as a background service. Open http://127.0.0.1:11434 in your browser — if it shows "Ollama is running" you're ready. If not, find Ollama in the Start Menu and launch it, or run ollama serve in a terminal.
• macOS / Linux: Run ollama serve if it isn't already running in the background.

Windows tip: If ollama serve returns bind: Only one usage of each socket address, Ollama is already running — skip this step entirely.

Step 2 — Pull a model (first time only)

Dotstell needs two model types — one for chat/writing and one for semantic search (RAG). Pull one from each column:

ollama pull gemma4:e4bollama pull nomic-embed-text

Models that think (Qwen 3.5, Gemma 4 and similar) work: dotstell turns the thinking off so replies are quick. You can allow it with Let the model think before answering under the model picker in AI Settings; expect much slower replies.

No GPU? Stick to 3–4B models (phi4-mini, qwen2.5:3b, llama3.2). 7B models are slow on CPU alone.

Step 3 — Start the Dotstell Local AI Agent

You do not need to clone this repo. Run directly via npx (Node.js 18+ required):

npx @dotstell/agent

Keep this terminal window open while you use Ollama in dotstell. The agent runs on http://127.0.0.1:12345 and is loopback-only — not reachable from outside your machine.

Step 4 — Open dotstell.app → AI Settings → choose Ollama (Local)

The settings modal shows a green "Local Agent is running" badge when the agent is detected. All AI features — Chat, Writing Assistant, Inline Assist, Summaries, Person Intelligence, AI Digest — route through the agent automatically.

Alternative ways to run the agent:

# Download the single file directly (no npm):curl -o dotstell-agent.mjs https://raw.githubusercontent.com/dotstell/dotstell/main/packages/agent/index.mjsnode dotstell-agent.mjs
# If you already have the repo cloned:node packages/agent/index.mjs

See packages/agent/README.md for environment variables, troubleshooting, and security details.

You do not need the agent when running dotstell locally (localhost:3000). PNA restrictions only apply when accessing the live https://dotstell.app.

Tech Stack

See docs/ARCHITECTURE.md for a full breakdown of the system design, data model, security model, and key design decisions.

Project Structure

The full annotated directory tree is in docs/ARCHITECTURE.md.

Roadmap

Contributing

Contributions are welcome and genuinely appreciated.

• Bug reports & feature requests — open an issue
• Code contributions — please open an issue first for anything significant so we can align on approach
• Docs, design, tests — always welcome without prior discussion

See CONTRIBUTING.md for full guidelines.

Self-Hosting

Dotstell is designed to be self-hostable. You need:

• A Node.js server (or Vercel / any Next.js host)
• A Supabase project (or self-hosted Supabase)

The full guide — database, environment, Supabase auth settings, running it, and what each feature needs — is in docs/DEPLOYMENT.md.

License

GNU Affero General Public License v3.0 — AGPL-3.0

This means you can use, modify and distribute Dotstell freely, including self-hosting it. If you run a modified version as a network service, you must make the modified source available to users of that service.

© 2026 Dotstell