Founder of Clearminutes. Privacy-first AI meeting assistant that transcribes, summarises, and extracts action items.

youtube

projects

devto

github

James Gibbard  • Peerlist

Clearminutes is the only polished, cross-platform, GPU-accelerated meeting assistant where your audio is architecturally incapable of leaving your machine. Transcription, summaries, and action items all run locally by default - not as a privacy policy you have to trust, but as a design constraint that can't be bypassed.<ul><li>Local by design, not by policy - Whisper runs on your hardware (Metal/CUDA/Vulkan). No audio is ever transmitted to any server.</li><li>Built-in AI model - Ships with Gemma for summaries, no API keys or accounts required. Optionally connect your own cloud provider (Claude, OpenAI, Groq, OpenRouter) if you want.</li><li>GPU-accelerated transcription - Metal on macOS, CUDA on NVIDIA, Vulkan on AMD/Intel. Not a cloud API bottleneck.</li><li>Cross-platform desktop app - Native macOS, Windows, and Linux support via Tauri + Rust, not an Electron wrapper.</li><li>Professional audio mixing - RMS-based ducking and VAD filtering produce clean recordings and accurate transcripts, even with system audio.</li><li>No bots in your meetings - Clearminutes records locally from your device. No meeting bot joins your call, no third party intercepts your audio, no one on the call even needs to know you're recording.</li><li>Zero cloud dependency - Otter, Fireflies, and Granola all send your audio to their servers. Clearminutes processes everything on-device. If the internet drops mid-meeting, it keeps working.</li><li>No subscription required for core features - Free tier includes full transcription, summaries, and meeting history with no time limits. Pro adds speaker identification, PDF export, and TickTick integration.</li><li>One-time lifetime option - No recurring billing, no vendor lock-in on your data.</li><li>Your data stays yours - All meetings stored in local SQLite. No cloud dashboard. No training on your conversations. You can verify this because the code architecture makes it impossible to phone home audio data.</li></ul>