A free, open-source, local-first dictation app for Mac: hold Option + Space, speak, and cleaned-up text is typed into any app — transcribed entirely on-device, with no account, no subscription, and no cloud by default.
Last verified · 2026-10-11 · by Moe Ameen
BetterWispr (betterwispr.com) is a free, open-source dictation app for macOS, published by the opennookorg project under the Apache 2.0 license and described openly as a local-first alternative to Wispr Flow. The interaction is the familiar one: hold Option + Space in any application, speak, release, and the cleaned-up text is typed straight into the field you were in — with a clipboard fallback if the app blocks a paste. It removes filler words and false starts, applies spellings you teach it, and matches a per-app writing style you set.
The defining trait is that it runs on your Mac. BetterWispr ships several on-device speech models — Apple's built-in speech, NVIDIA's Parakeet (v3 for 25 European languages, v2 for English, plus a Japanese model), and Whisper Large v3 Turbo via WhisperKit — that download once and then transcribe offline. There is no account, no subscription, and no cloud transcription fallback; the project's phrasing is "speech stays on your Mac." You can optionally connect Sarvam AI, Smallest AI, or any OpenAI-compatible endpoint with your own key, and a couple of optional features (Medium cleanup and meeting summaries) can route text through a cloud model or a local Ollama install — those are the only paths that leave the machine, and they are opt-in.
Around the core sit cleanup levels (Light by default, Medium for clarity, None to keep every word), spoken punctuation and "new paragraph" / "scratch that" voice commands, a learned custom vocabulary, a history that stores both raw and corrected text with a "Use Original" restore, usage insights, and a meeting notetaker that transcribes your mic and (on recent macOS) other apps' audio. It requires macOS 14 or later, with Apple Silicon recommended for local model performance, and ships separate Apple Silicon and Intel builds. Cleanup, styles, and voice commands work in English; model-dependent transcription covers more languages.
The honest boundaries: BetterWispr is Mac-only, with no Windows or mobile apps, and it is a small open-source project with a single primary contributor rather than a funded product — capable, but community-supported. And like every dictation tool, it is an input layer: it produces clean text and stops there. It does not generate video, images, carousels, or persona footage; it does not caption or reframe media per platform; and it does not schedule or publish anything.
Think of BetterWispr as your private microphone and [Kompozy](/) as the public studio it feeds. BetterWispr's job ends the moment you have clean, on-device text; Kompozy's begins exactly there, and it needs precisely the kind of input BetterWispr is best at producing. Because the cleanup level and voice commands already strip the "ums," add punctuation, and apply your custom spellings, what lands in your clipboard is a tidy script rather than a raw transcript — which is the ideal source for Kompozy's generators. Paste it in and Kompozy turns one dictated script into a [Persona Short](/glossary/persona-shorts) or [Persona Frames](/glossary/persona-frames) clip where your face-locked avatar reads it to camera, captioned and sized 9:16 — the talking-head media a dictation app has no way to make. The private thing you spoke on your laptop becomes a published video without a shoot.
From that same dictated script, Kompozy fans out the rest of the week: a brand-exact [Carousel](/glossary/hyperframes), [Quote Graphics](/glossary/output-buckets) pulling your strongest lines, native Text Posts, a Blog Article, and an Email Newsletter — every one governed by the [Persona Brief](/glossary/persona-brief) so a casual spoken riff comes out in one consistent brand voice with a banned-word filter applied. Then Kompozy schedules and publishes the full set across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot), behind a per-post review gate. BetterWispr keeps capture free and off the cloud; Kompozy makes the video, images, and persona formats it can't, and does the publishing it doesn't.
BetterWispr is a free, open-source, local-first dictation app for macOS. You hold Option + Space, speak, and cleaned-up text is typed into whatever app you are in. Transcription runs on-device using Apple, Parakeet, and Whisper models, with no account, no subscription, and no cloud by default. It is published by the opennookorg project under the Apache 2.0 license as an alternative to Wispr Flow.
Yes to both. It is free and open-source (Apache 2.0) with no subscription or account, and its built-in speech models download once and then transcribe entirely on your Mac, so dictation works with no internet. The network is used only to download a model or if you opt into an optional cloud/API feature such as Medium cleanup or meeting summaries.
It is Mac-only (macOS 14+, Apple Silicon recommended), with no Windows or mobile apps, and it is a small single-contributor open-source project rather than a funded product with a guaranteed roadmap. It is also an input tool: it produces clean text and does not generate video, images, or carousels, and it does not caption, schedule, or publish.
No. It produces clean text only. To turn dictated words into persona video, carousels, quote cards, blogs, newsletters, and scheduled posts, you pair it with a content engine. Kompozy accepts the pasted text as a source and generates and publishes finished content from it across nine platforms.
BetterWispr is free, open-source, local-first, and Mac-only — transcription stays on your device by default. Wispr Flow is a funded, paid product with a cloud backend and Mac, Windows, iOS, and Android apps. BetterWispr trades cross-platform reach and a vendor roadmap for privacy, zero cost, and open code.