PromptStream is system-wide AI dictation for macOS and Linux. Hold a key, say what you mean, release — and about a second later, clean text lands wherever your cursor is. Not in a special window you copy-paste from: in the app you were already using. Slack, Gmail, VS Code, your terminal, a browser form, an AI chat.
The pitch is simple: you speak around 150 words a minute and type around 40. PromptStream closes that gap and keeps the part of dictation that used to ruin it — the messy transcript — out of your way. Filler words, false starts, and mid-sentence corrections are cleaned up before the text is typed. What appears is what you meant, not what you said.
Now on Linux, too
PromptStream started on macOS, and as of this release it runs on Linux as well — Fedora, Ubuntu, and most other distros, as a single AppImage you download and run. Same hold-to-talk hotkey (Right ⌘ on a Mac, Right Super on Linux, remappable to anything), same overlay pill with a live waveform, same cleanup. If you live in a terminal and a browser on a Linux machine, dictation is no longer a Mac-only luxury.
Prompting AI, faster than you can type it
The use case that surprised us most: talking to AI. Tools like Claude, ChatGPT, Cursor, and Copilot reward long, detailed prompts — the more context you give, the better the answer. But nobody enjoys typing four paragraphs of context into a chat box, so most people type two sentences and get a mediocre reply.
Speaking flips that economy. Describing what you want out loud — the background, the constraints, the edge cases, the tone you’re after — takes a fraction of the time, so you actually do it. Hold the key, think out loud for thirty seconds, release: a clean, detailed prompt lands in the chat box, ready to send. Prompting stops being a typing task and becomes a talking one, which is what it always wanted to be.
This works anywhere you prompt: web chats, Cursor’s composer, a CLI agent running in your terminal. Wherever the cursor blinks, your voice types.
The everyday cases
The rest of the day benefits the same way. Email replies that sat in your inbox because typing them felt like work. Slack messages spoken in the time it takes to switch channels. Doc drafts, commit messages, code review comments, meeting follow-ups — anything you’d say out loud anyway, said out loud once and typed for you. The math on the speed gap is in speaking vs. typing: if you write two hours of prose a day, speaking the first drafts gives most of it back.
And if typing hurts — RSI, wrist pain, long keyboard days — dictation isn’t a productivity trick, it’s relief.
How it works
Hold your hotkey and a small pill appears with a live waveform, so you know it’s hearing you. Speak naturally — ums, restarts, and “wait, scratch that” are all fine. Release, and your speech is transcribed, cleaned by an AI pass that fixes punctuation and drops the mess without rewriting your meaning, and pasted at your cursor. Around a second, end to end. Your dictation history stays on your machine, audio is processed transiently and never stored, and if anything fails mid-flight your raw words are kept on the clipboard — dictation never fails silently.
PromptStream is free for 5,000 words a month — no credit card, nothing to configure. If it sticks, Pro is $7 a month or $59 a year for unlimited dictation, less than half of what Wispr Flow charges. Download it for macOS or Linux and say your next email instead of typing it.