Think out loud, keep the result
Talk through an approach and end with a linked note rather than a lost train of thought.
For developers
Prompt out loud. Keep every design decision.
Speak the prompt instead of typing it. Record the design review without a bot. Ask later why the schema changed, and get the line that says so. Native on macOS, Windows and Linux.
- Terminal syntax left alone
- Native on Linux, Wayland too
- MCP for Cursor and Claude
- No bot in standups
For developers
A day with Mr.Speech
- 09:30Standup, no bot
Recorded on your laptop. Action items land on the board with owners and due dates.
- 11:00A prompt with real context
Two minutes spoken into Claude Code: the bug, the constraint, the edge case. Typed where your cursor is.
- 16:30PR description, spoken
Dictated in the browser, then shortened with one ask. Library names spelled right from your dictionary.
- Native on Linux. Real Wayland, Hyprland and KDE support, a .deb and an .rpm, alongside macOS and Windows.
- It knows a terminal from a document. Syntax untouched in a shell, structured prose in an editor, a casual line in chat. It detects the app you are dictating into.
- Your notes and boards, reachable from your tools. Claude Code, Cursor and ChatGPT read and write your workspace. Ask one to pull yesterday's decisions into a board, and it can.
For developers
How it fits the day
01
Prompts with the whole context
Press the hotkey and talk through the bug, the constraints and the edge cases. The text lands in your editor, your chat or your agent's prompt. Save your best prompt as a snippet and say its name to drop it in. In a terminal, syntax is left alone.
02
Explain it without touching it
Select an unfamiliar term in a spec and ask Mr.Speech to explain it. Your text stays as it is. Or select a rambling PR description and ask for it shorter, or as bullet points.
03
Design reviews without a bot
Both sides of the call are recorded on your own machine, whatever the meeting app. Speakers are labelled. One click gives decisions, action items and discussion points. The AI is instructed never to invent a decision, task or name.
04
Why was it built like that?
File the review into the feature's project with one move. Weeks later, ask Mr. AI what was decided about the cache. It answers from every transcript and note in the project and links to the exact line.
05
Linux native, agents connected
Runs natively on macOS, Windows and Linux, including Wayland, Hyprland and KDE. Over MCP, ChatGPT, Claude, Cursor and Claude Code can read and write your notes, tasks and boards.
For developersDesign Review Agents
Where it earns its place
Long review comments
Say the reasoning once; it lands in the review box formatted for that box.
Your workspace as a data layer
Ask your assistant to pull yesterday's decisions into a board, and it can, because the connector writes.
Questions
Questions
Will it mangle my code or my commands?
Formatting follows the app you are in. In a terminal, syntax is left alone, and a literal cleanup style keeps exactly what you said. Library and tool names go into your personal dictionary, and your corrections come back as one-click suggestions.
Does it work on Linux with Wayland?
Yes. Mr.Speech runs natively on macOS, Windows and Linux, including Wayland, Hyprland and KDE. The same hotkey dictates into your editor, your terminal and your browser.
Can I run it without internet?
Yes, in the fully private mode. Dictation, notes, search and the transcription of recorded meetings then run on your machine. When you choose the cloud, it runs on European infrastructure.
