Canto is a desktop notebook app powered entirely by local AI — everything runs on your device via native Metal GPU acceleration, so your notes never leave your machine.
Unlike Notion AI or Obsidian Copilot, Canto requires no API keys, no subscriptions, and no internet connection after the initial model download. All AI inference runs locally via node-llama-cpp.
Key Features:
• Memory Links — A personal knowledge graph that semantically connects your notes using MiniLM embeddings. The AI reads your entire notebook locally and surfaces related ideas you've forgotten.
• Instant AI Autocomplete — GitHub Copilot-style inline ghost text suggestions as you type, with sub-200ms latency because there's no network roundtrip.
• AI Agent Panel — A full conversational agent with tool-calling for reading, editing, creating, and navigating your vault.
• Vault Chat — Chat with your entire knowledge base. AI queries across all your notes using RAG to answer questions about your own writing.
• Writing Style Profile — AI learns your writing voice from your notes and generates text that sounds like you.
• Rich Editor — WikiLinks [[Note]], collapsible toggle blocks, live Table of Contents, Mermaid diagrams, LaTeX math (KaTeX), Notion-style table controls, code execution (Jupyter-style notebooks with Python + TypeScript), file attachments, and slash commands.
• Version History — Auto-saves up to 50 versions per note or notebook with diff preview and restore.
• Encrypted Storage — AES-256 encrypted SQLite database. Your data stays private even on disk.
• Import / Export — Supports .canto (encrypted note), .cwsp (encrypted workspace), Markdown, HTML, Evernote ENEX, DOCX, and PDF.
• External AI Endpoints — Connect to Ollama, LMStudio, or any OpenAI-compatible endpoint in addition to built-in local models.
Built-in AI Models (GGUF, run 100% locally): SmolLM3 3B, Qwen 3 4B/8B/14B/32B, GPT-OSS 20B, Qwen3 Coder 30B, Qwen 3 Next 80B.
Currently available for macOS (Apple Silicon). Windows and Linux support planned.