Official documentation
HoldSpeak README (GitHub repository)
karolswdev, GitHub (karolswdev/HoldSpeak), 2026
Read the originalRead on October 25, 2026
- What it is
- The README of the HoldSpeak repository, published by its maker. It describes local Whisper transcription, typing into any app, optional rewriting with the user's own model, and that there is no HoldSpeak cloud.
- How much weight it carries
- The maker's own repository README, read from the default branch on the date above. The project labels itself pre-alpha. It states no pricing or retention terms; the security document in the repository was not used here.
Facts taken from this source
- The HoldSpeak README describes the app as everything local, open source, with no subscription.
“Strong accuracy and editing UX, no model management | Everything local, open source, no subscription, meetings” (Compare)
- The HoldSpeak README says there is no account and no HoldSpeak cloud.
“There is no account and no HoldSpeak cloud.” (Headline section)
- The HoldSpeak README says the dictation pipeline can rewrite a rough prompt with the user's model before it types.
“adds your project's facts and context, and can rewrite a rough prompt with your model before it types” (Speak anywhere)
- The HoldSpeak README says transcription is always local.
“Transcription is always local. Model work goes to the host you assign in” (How it works)
- The HoldSpeak README says the language model can be any OpenAI-compatible endpoint the user chooses.
“The language model is yours: a local GGUF or MLX model, or any OpenAI-compatible endpoint you choose.” (Headline section)
- The HoldSpeak README says that with the hotkey held, the dictated text lands in any app.
“hold a key, speak, the text lands in any app and improves with each correction” (What's inside)
- The HoldSpeak README describes the app as all local, with Whisper running on the user's machine.
“Hold a key, speak, it types. Record a meeting, it closes the loop. Ask anything about your work. All local.” (Headline)