Local ASR Model Butler

A local transcription CLI/GUI that manages models, parameters, and punctuation for you

Problem

People who want on-device transcription hit the same three walls: cloud ASR bills by the minute and uploads private audio; whisper.cpp means hand-compiling flags, model formats, VAD chunking and long-audio stitching, repeated on every new machine; YouTube/Bilibili captions only cover some videos and lack punctuation. One user already built a 76-model CLI plus a note-pipeline agent skill, proving demand but also that the barrier is still too high.

Solution

A cross-platform desktop app: one-click download and management of multiple local ASR models (including quantized ones), automatic hardware detection and model/parameter recommendation, built-in VAD chunking and long-audio stitching, Chinese sentence segmentation and punctuation restoration, and Markdown output ready for Obsidian/Notion. Batch folder processing, speaker diarization, and a hotkey 'transcribe to clipboard/notes'.

Users

Podcast and meeting-note power users, content creators, Chinese-language users, privacy-sensitive researchers and journalists

MVP

Model download with hardware-based recommendation; drag-and-drop batch transcription with auto punctuation; Markdown/subtitle export into a notes library; clipboard hotkey transcription; CPU/GPU benchmark comparison table

Revenue

Free CLI plus paid desktop GUI one-time license (~$29); optional curated model packs; offline enterprise deployment license

Why now

Local ASR quality is finally good enough, but the ecosystem is still command-line assembly; giants won't build the 'help you run your own model' UX layer, which suits an indie.