Typilot

• Meetings. One tap records your mic and the system audio. You get a live transcript, speaker labels, a summary with decisions and action items, and a chat to ask questions across your meetings.
• Dictation. Speak into any app, with automatic punctuation and spoken commands like "new line".
• Writing help. Inline autocomplete and rewriting work system-wide. Typing @screen adds what's on your screen as context.
• The rest. Notes, local chat and clipboard history. Audio, transcripts and notes stay on the device.

• Meetings without headphones. The mic picks up the call from your speakers, so other people's words end up under "You". Typilot runs WebRTC AEC3 with the system audio as reference, adds noise suppression, mutes the mic in spans that are only echo, and filters repeats out of the transcript.
• Knowing who spoke. Each remote voice is separated and labelled, your own voice is "You", and optional voiceprints recognise recurring speakers across meetings.
• Reading the screen on three systems. Text recognition is built in with no extra install, using macOS Vision, Windows.Media.Ocr and ocrs on Linux.
• Shipping a desktop app. Installers for all three systems with code signing and auto-update, and a backend for trials, Stripe subscriptions and Ed25519-signed entitlements.

Tauri 2 with a Rust core, React and TypeScript for the interface, SQLite for local storage. Speech runs on sherpa-onnx and Whisper, language models through Ollama. The licensing backend is Node.js and Supabase, and the website is Next.js.

Around 150 signups and 6 paying customers. Those are small numbers, but they are real people paying for something I shipped and still maintain.