Knovum 0.0.2: Building an Open-Source Local-First AI Canvas with Tauri 2, Vue 3, and Rust

Description and use

Knovum 0.0.2 — an open‑source local AI workspace with an infinite 2D canvas, 45 MB idle RAM, and local-first storage, built on Tauri 2, Vue 3, Rust, SQLite, and LanceDB.

Problem

Existing desktop AI wrappers suffer from a linear chat interface that hampers deep research synthesis and consume 700 MB–1 GB of RAM before loading model weights.

Outcome

Knovum achieves a local‑first AI workspace with sub‑second startup, 45 MB idle RAM, and a 2D canvas that supports side‑by‑side document comparison, while maintaining efficient vector search and low memory usage.

Workflow

• 1. Use Tauri 2 to orchestrate a native WebView with a compiled Rust backend for sub‑second cold boot and 45 MB idle RAM.
• 2. Build a Vue 3 frontend with Konva for an infinite 2D canvas and JSON Canvas specification for workspace persistence.
• 3. Store notes, boards, and relations in SQLite with strict migrations.
• 4. Run an in‑process LanceDB for fast vector similarity search without external daemons.
• 5. Route local LLM calls through Ollama API and embed Whisper for on‑device transcription.
• 6. Implement differential space indexing to rebuild only gaps during updates.
• 7. Enforce streaming conversation boundaries and mutation‑aware citation integrity.
• 8. Harden Windows desktop with opaque WebView2 bootstrapper and safe async window reveal.

Evidence

Existing desktop frontends repeatedly hit two frustrating walls: The 1D Chat Trap and The Electron Memory Tax.

Source

Knovum cold‑boots in sub‑second time and idles around 45 MB of RAM.

Source

Every visual workspace is stored directly on your local filesystem using the open JSON Canvas (.canvas) specification.

Source