Speechmark overview

Speechmark is a macOS meeting-memory app that records straight from system audio (no bot joins the call), then transcribes and diarizes fully on-device. Unlike most tools, it keeps the original audio — so you can replay and verify exact wording. Summarize locally with Apple Intelligence or Ollama, or bring your own Claude/OpenAI/Gemini key (transcript text only). A Claude Desktop MCP connector lets you ask questions across your meeting history while everything stays on your Mac.

What's new in version 1.3.0

Catch me up. Drifted off? While recording with live transcription, press Catch me up in the recording bar for 3–5 bullets on the last 10 minutes (or 5, 15, 30, or the whole call so far), using your summary provider. Nothing is saved; the card clears when the recording stops.
Hear when someone says your name. A notification shows the sentence your first name came up in. Click it to jump to that line. Turn it off in Settings if you don’t want it.
Bookmark important moments. Press ⌃⌥⌘B (or the bookmark button) during a call to mark the moment; you can change the shortcut in Settings › Recording. Bookmarks show up in the transcript, and your notes are written to cover every one of them.
“Last time” when a meeting starts. When a calendar meeting starts, you see what came out of your previous meeting with the same people: summary, decisions and open action items. The start notification also shows how many there are.
Live transcript shows who’s speaking. On headphones, live lines are labelled You or Them.
The live transcript is never lost. It’s saved to disk as the call goes, so a crash mid-call no longer loses it. And if the full transcription after a call comes back empty, the live transcript is kept, with a Retry button, instead of being wiped.
Clearer speaker numbers. Speakers Speechmark can’t identify are numbered within each meeting (Speaker 1, Speaker 2), not “Speaker 85”.
Much lighter while recording. Recording no longer keeps a whole CPU core busy for the length of the call — the live waveform was redrawing far more than it needed to. Your Mac should stay cooler and your battery last longer during calls.
Transcripts line up after a headset switch. If a Bluetooth headset connected or disconnected early in a call, other people’s lines could land seconds too early, and some of their words were credited to you. They’re now placed at the right time, and playback, the waveform and trimming follow the same timeline. Re-transcribe an older meeting to fix it there too.
Newer Claude models work. Speechmark works with the latest Claude models, and there’s a new Workspace ID field for Anthropic keys that cover more than one workspace.
More for Claude Desktop. Claude can now answer questions about the meeting you’re recording right now, read just one part of a transcript (one speaker, a time range or a keyword), brief you before your next meeting, and see who spoke in each meeting. To get these, open Settings › Assistant and click Add to Claude Desktop again.
Setup no longer crashes with some Ollama models. If you use Ollama and have two sizes of the same model installed (for example qwen3:8b and qwen3:4b), Speechmark could quit a few seconds into first-time setup. It now picks one of them and carries on.

Speechmark for Mac

Speechmark overview

What's new in version 1.3.0

Speechmark for Mac

Write your thoughts in our old-fashioned comment

(0 Reviews of Speechmark)

How would you rate Speechmark?