Talk faster
than you type.

Terminal, editor, browser, chat box. Real keystrokes go to the focused window, nothing to copy or paste.

Run the fully offline path (Local Whisper plus a local Ollama model) and nothing leaves your machine. No account, no telemetry, no servers of ours. Open source, so verify it.

Dictate prompts straight into Claude Code, Cursor, or any terminal, 3 to 5 times faster than typing them. Talk the way you think.

Install once and you're done. PipeVoice checks for new versions in the background, verifies each one with SHA-256, and updates itself. Open the About window any time to see what's new or update on the spot. No uninstall, no reinstall.

A quiet tray app, not a memory hog. No Electron, no background bloat. It waits in the tray until you hold the key.

A hotkey for every Voice plus a picker, type or clipboard output, accent and language picker, free-text speech notes, vocabulary boosting and word-fixes for your jargon, push-to-talk or toggle, paste-or-type output, plus min-recording, paste-speed and Deepgram-timeout controls. Open source, so fork it.

Say "new line", "new paragraph" or "tab key" to format as you go. Say "scratch that" to discard the last bit, or end with "send it" to type and press Enter hands-free. Toggle it on or off.

Every transcript is saved locally on your PC. Open the tray History window to re-copy anything you dictated earlier, marked typed or clipboard. Stored only on your machine, nothing uploaded.

Per-app profiles. PipeVoice spots the focused window and switches behaviour to match: raw text and a hands-free Enter in your terminal, polished and auto-sent in chat, no cleanup in your editor. Set the engine, cleanup, Enter and output once per app.

A hotkey starts and stops it. Your mic and the call's audio are recorded separately, so you are never mistaken for the other person. Remote speakers are still separated, and your dictation hotkey switches itself off while recording so a stray press can't interrupt.

Click a "Them 1" label and it plays the clearest two seconds of that person, so you can recognise them and type their name. Names never overwrite the original transcript.

Mark a moment just by saying so — no key, no hands. Nothing listens live; the phrase is found when the recording is transcribed, so it costs nothing and works offline. A bookmark marks the half-minute before it, because you always react after the interesting bit.

Turn a meeting into bullet points, to-dos, or actions with an owner and a deadline. Moments you bookmarked get priority. It won't invent owners or dates nobody agreed to.

Right-click any word in a transcript and correct it. It changes every occurrence — and if you like, the same fix applies to your normal typing from then on. So a name it keeps mishearing gets fixed once, not every time.

Find a word across all your recordings, not just the one you have open — with a match count and a snippet of where it came up. It searches what you see, so a name you have corrected is findable by its corrected spelling.

One press strips "um", "uh" and false starts and fixes punctuation. It only tidies — it never rewords anyone or changes what was said — and the raw transcript stays one click away.

Each person's share of the conversation, how many turns they took, how many questions they asked, and the longest stretch nobody interrupted. Worked out from the transcript on your own machine, so it costs nothing and works offline.

Optional. PipeFocus watches the conversation live and speaks up only when something concrete is going wrong — an action item nobody has taken, a decision deferred again. At most one nudge every few minutes, and silent when nothing is wrong. Needs Deepgram, and costs about 30¢ an hour of meeting.

Export a meeting as Markdown, a self-contained web page you can print straight to PDF, or slides. Everything you see comes with it — highlights, notes and named speakers.