Open Wispr

Open Wispr sits in a narrow but valuable niche: pure, local, push-to-talk dictation for macOS at zero cost. Its biggest strengths are its privacy posture and its simplicity. Every piece of audio is recorded to a temp file, transcribed on-device via whisper.cpp with Metal acceleration, and deleted immediately — zero network requests, no accounts, no telemetry. That's a genuine differentiator in a market where competitors like Wispr Flow ($15/mo) send audio to the cloud and Superwhisper ($8.49/mo) is local but closed-source and paid. Open Wispr's MIT license is also the most permissive in the category, so you can fork it or audit the code if you care about exactly what runs on your machine. For the target user — a developer on Apple Silicon who wants to dictate long prompts into Claude Code or a terminal, or a writer who wants distraction-free voice typing — the experience is solid. The hold-to-talk workflow with the Globe key is fast and unobtrusive, text lands at your cursor in any app, and the menu bar icon with waveform feedback makes it clear when you're recording. The fact that it's a lightweight background service installed via Homebrew, with no Electron and no app bundle, keeps it out of the way. You can also switch to toggle mode (press once to start, again to stop) and choose from six model sizes, from tiny.en (75 MB, fastest) up to large-v3 (3 GB, best accuracy on M1 Pro+), plus quantized variants that cut size by a third to half with minimal quality loss. Configuration is where the trade-offs appear. Everything lives in a single JSON file, ~/.config/open-wispr/config.json, with a small set of options: hotkey, model size, language, spoken punctuation, maxRecordings, and toggleMode. This is powerful for terminal users but a real barrier if you're not comfortable editing JSON or running brew services restart open-wispr to apply changes. There's no graphical settings panel, and auto-updates don't exist — you'll need to run brew upgrade manually. The default model is base.en (142 MB), which is a good balance of speed and accuracy for English, but if you need another language you must switch to a non-.en model and set an ISO 639-1 code. The limitations are also hard boundaries: it's macOS-only, requires Apple Silicon (M1 or later), and offers no AI rewriting, formatting, or summarization features — that's by design, but it means you won't get the cloud-assisted polish that Wispr Flow or VoiceInk provide. For very noisy environments or strong accents, accuracy depends on which Whisper model you pick; the larger models (medium.en, large-v3) are meaningfully better but need more RAM and disk. Finally, the project is maintained by an individual (human37) on GitHub — the code is open and MIT-licensed, but there's no formal support channel or SLA, so you're relying on community goodwill and the installer script. Where it fits: single developers or students on Apple Silicon who want free, private, offline dictation and are comfortable with a