Hold a key, speak, release.
Clean written text lands in whatever app you're using. Transcription can run entirely on your Mac — recorded audio never has to leave it.
Spoken in, written out
um so basically can we uh move the standup to like ten thirty tomorrow because i have a a dentist thing in the morning
Can we move the standup to 10:30 tomorrow? I have a dentist appointment in the morning.
The cleanup pass removes fillers and false starts, fixes punctuation, and keeps your actual wording — it is not a rewrite. Provider is swappable (Gemini, OpenRouter, OpenAI, Claude, or a custom endpoint), it can be switched off entirely, and any failure falls back to the raw transcript instead of failing the dictation.
Why Plainsay
Open source, transparent, safe
Trust isn't a claim here — it's something you can go read. The whole client is MIT licensed and public: the pipeline, the HUD, the settings, all of it. Not open core, and not "open now, closed later" — free to fork, modify, and self-host indefinitely.
The entire dictation client — transcription, editing, the HUD, Settings, the Setup Assistant. On-device, Plainsay makes no network call you didn't ask for: no analytics, no telemetry, no phoning home.
The billing backend for the paid Cloud tier — it holds payment and provider credentials, which is exactly why it stays separate and private. The dictation experience you actually use is fully open.
github.com/conrader/plainsay — read every line
Three ways to run it
Plainsay Cloud is the easiest way to start — no model download, nothing to configure. Prefer everything on your Mac? Local is free and fully capable, one extra setup step.
Reliability
A dictation is never silently lost.
Falls back to the raw transcript automatically, with a HUD notice.
Text stays on the clipboard, with a HUD prompt to press ⌘V.
Audio is staged to disk before transcription, and recovered on next launch.
The dictation is already in local history — written there before insertion was attempted.
Coming
Full live typing — words landing in your document as you speak. Today's HUD live preview is the deliberate first step; true live typing trades away the "clean, then insert once" guarantee, so it ships separately.
Also: continued work on language auto-detection, and default-model changes as local models improve.
Start dictating in about two minutes.
The Setup Assistant walks you through everything on first run.