Why
Transcription services want the file uploaded and a monthly fee. A laptop with Whisper does the same job and the audio never leaves it. Subtext is the plumbing around that: a URL in from the phone, a transcript back, and a few preset prompts run against a local model.
What’s inside
- Transcribe a pasted URL, several URLs combined into one transcript, or an uploaded file.
- YouTube captions are used first; Whisper runs only when there are none.
- Download the original video or an audio-only stream.
- Preset analyses through Ollama: caption ideas, hook rewrites, title packs, or a custom prompt.
- Optional local speaker labels.
- Private access over Tailscale Serve, bound to localhost, behind a shared key.
- A CLI client and a desktop app with transcript editing and export to JSON, Markdown, HTML, PDF or text.
Status
Active. Whisper runs through openai-whisper, faster-whisper or mlx-whisper depending on the machine; the default model is large-v3-turbo.