Cadent

A solo project. Hold a hotkey, speak, release — clean text appears at the cursor in whatever app has focus. Cadent captures the mic, transcribes on-device with faster-whisper (biased toward a custom vocabulary, with sounds-like correction and snippet expansion), optionally runs the transcript through a local llama.cpp model to clean it up behind a hallucination guard, then injects the result via SendInput or a clipboard paste and logs it to a searchable SQLite history. Everything runs on the machine — the only network traffic is the explicit first-run model download. Built in Python with a per-OS platform seam behind a single factory, GPU acceleration that proves itself with real work before committing (CUDA and Vulkan on Windows, Metal on Apple silicon) and falls back to the CPU when it cannot, and shipped as a Windows installer and an Apple-silicon disk image.

Stack

• Python
• Windows
• Apple
• Database