Knovy started as an internal startup project in 2025. Over time, we stripped out all cloud infrastructure, removed the login gate, and made it fully local and open source. Everything runs on your machine — no accounts, no API keys, no telemetry. This will be a training ground project for the agentic tech stack.
Knovy is a local-first AI desktop assistant for real-time audio analysis, transcription, and AI-powered interactions. It runs fully on your machine — no cloud backend, no account, no API keys required.
Compact floating bar with auto-update notification
Knovy is a single cross-platform Electron desktop application that combines local whisper.cpp transcription with local Ollama AI actions to provide intelligent assistance during meetings and conversations:
- Dual-stream audio capture (microphone + system audio)
- Offline speech-to-text via bundled whisper.cpp
- AI actions (summarize, chat, keyword search, screenshot analysis, deep response,
recommendations) powered by a local Ollama server (
http://localhost:11434) - All persistence in a local SQLite database
There is no cloud backend. Earlier web, admin-dashboard, and Supabase components have been removed; the repository is now a single-package Electron app.
A single-package Electron + Vite desktop application rooted at the repository root, managed with pnpm.
/
├── src/ # Electron app source
│ ├── main/ # Main process (window mgmt, IPC, SQLite)
│ ├── renderer/ # React UI (renderer process)
│ └── preload/ # Secure IPC bridge
├── resources/ # Bundled binaries (whisper.cpp, models)
├── code-signing/ # macOS signing / notarization scripts
├── tests/ # Vitest tests
├── electron.vite.config.ts # electron-vite build config
├── electron-builder.yml # Packaging / publish config
└── package.json # Single root manifest
- Node.js v20 or later
- pnpm v10 or later
- Git
- (Optional) Ollama running locally for AI actions
-
Clone the repository
git clone https://github.com/Intevia-AI/Knovy.git cd Knovy -
Install dependencies
pnpm install
-
(Optional) Set up environment variables
The app runs without any cloud credentials. If you need to override defaults, copy
.env.exampleto.env. -
Start development
pnpm dev
The app will automatically download the small whisper model (~488MB) on first launch.
- Local Transcription: Privacy-focused speech-to-text using whisper.cpp (runs offline)
- Dual-Stream Audio: Simultaneous microphone and system audio capture
- Two-Stage Language Detection: Improved accuracy for Traditional Chinese
- Progressive Enhancement: Raw transcription displayed immediately, AI-enhanced version follows
- AI Actions: Summarize, chat, keyword search, screenshot analysis, deep response, recommendations — all via local Ollama
- Chinese Language Support: Automatic Traditional/Simplified conversion via OpenCC
- Fully Local: No backend; SQLite for storage, whisper.cpp for transcription, Ollama for AI
- Progressive Enhancement: ID-based updates prevent duplicate transcriptions
- Chinese Language Support: OpenCC integration for Traditional ↔ Simplified conversion
- Release Management: Automated builds and updates via GitHub Actions CI/CD
Desktop app releases are automated via GitHub Actions and published to this repository's Releases.
- Update version in
package.json - Create and push a git tag matching
v*.*.*:git tag v0.3.9 git push origin v0.3.9
- The Release workflow builds, signs (macOS), and publishes the release.
- The desktop app automatically notifies users of new versions.
Code Signing: Requires Apple Developer credentials in repository secrets (see .github/workflows/release.yml).
pnpm install # Install dependencies
pnpm dev # Start the desktop app in development
pnpm build:local # Build locally (unsigned)
pnpm build # Build the signed macOS app
pnpm format # Format code with Prettier
pnpm test:run # Run config/release tests (Vitest)
Session history — browse and manage past recordings
Live recording view with real-time transcription and AI action panel
See CONTRIBUTING.md for development workflow, coding standards, and pull request process.