Speechmark

Speechmark is a private, on-device AI meeting-notes app for macOS. It records, transcribes, and summarizes meetings entirely on your Mac, turning a conversation into decisions, action items, and a short recap written in plain editorial prose, alongside a speaker-attributed transcript. It is built for people whose meetings matter — product managers, legal and compliance teams, consultants, and founders and executives — and it lives quietly in your menu bar. There is no account, no login, and no bot joining the call. Most meeting recorders are cloud-first. Otter.ai and Fireflies upload your audio to their servers and charge per seat, and a recording bot typically has to join the call. Granola transcribes on-device, but it syncs your notes to its cloud and, unless you opt out, uses anonymized meeting data to improve its AI models. Speechmark was built as the private alternative: audio, transcript, and notes stay on your Mac, no bot is dialed into your call, no account is required, and it is a one-time purchase rather than a subscription. The problem it addresses is familiar to anyone who leaves a meeting with pages of bullet points and no clear record of what was actually agreed. One early-access user, a director of product, described leaving meetings with three pages of bullet points and no idea what had actually been agreed, and now leaving with a paragraph that reads like a memo from someone who took it seriously. Speechmark's approach is deliberately narrow: three things, done well, and nothing else. The first is capture. The app lives in your Mac's menu bar, and one click starts recording. It captures your microphone and the meeting audio together and handles the routing for you, so there is no browser extension to install and no bot that joins the call. Because nothing joins the meeting on your behalf, other participants see no recording bot, and you do not have to change how you join or host calls. That single-click capture is the entire recording interface — there is no complex setup and no additional configuration required before you start. The second step is listening. Speechmark separates speakers automatically and labels them — the first time you confirm a name, that speaker is identified, and after that every transcript reads like a script rather than an undifferentiated wall of text. The result is a speaker-attributed transcript in which you can see who said what. This matters for meetings where attribution carries weight: knowing that a decision came from a specific stakeholder, or that a particular action item was accepted by a named person, is often as important as the words themselves. Every line in the transcript is also clickable, so you can jump to the exact moment something was said and review it in context without scrubbing through a recording. The third step is writing. Speechmark writes the note you would have written: decisions, action items, and a short recap, all in plain editorial prose rather than a bulleted dump. You choose which model generates that note, so you can pick the one that fits your privacy bar. The options stated on the site are Apple Intelligence, a local Ollama model, or your own OpenAI or Anthropic key. If you choose a cloud model with your own key, only the transcript text is sent — never the audio — and if you prefer, you can stay fully on-device with Apple Foundation Models or a local Ollama model. That model choice is what makes the note-generation step configurable to different privacy requirements while keeping the recording and transcription stages local. Speechmark also connects to Claude. From Settings → Assistant → Add to Claude Desktop, it installs as a Claude extension in one click — no config files, no terminal, and no API key required. Once connected, your transcript folder becomes a memory that Claude can reason over, privately and on your device. Claude answers from your own notes — decisions, action items, attendees — and cites the meeting each answer came from. The connector reads your notes locally and makes no network calls of its own, so you decide what any question surfaces. The site illustrates this with a query about pricing across a set of Acme calls, which returns the decisions reached across three meetings (annual billing at a 15% discount, with the per-seat floor still open), the two unresolved follow-ups, and the source meeting for each item. Product Hunt describes this as integrating well with Claude Code using an MCP server. Privacy is described as architectural rather than a setting. Recording, transcription, and note generation all happen on your Mac. Audio never leaves the machine — it is never stored or streamed to the cloud. Transcription runs locally on Apple silicon, and notes are generated on-device by default. The only thing that can ever leave is transcript text, and only as an opt-in choice you make yourself, for example if you select a cloud AI model for the summary using your own OpenAI or Anthropic key. The site lists the guarantees plainly: recordings and transcripts stay on your Mac, transcription runs on-device, no bot is dialed into your call, there is no account, login, or sign-up, usage stats are opt-in and off by default, and a cloud model sends text only, never audio. The practical benefit is that you leave a meeting with a document you can use. Instead of a wall of bullet points, you get a paragraph that reads like a memo, with decisions and action items already separated out and attributed. Because Speechmark tracks who said what and when, sprint retros, PRD updates, and stakeholder readouts practically write themselves, and you can stay in the conversation instead of racing to keep notes. The clickable transcript means you can verify any statement against the exact moment it was made. And because the AI assistant layer is grounded in your own notes with citations back to source meetings, you can ask questions across many meetings and get answers that point to where they came from. Speechmark is aimed at people whose meetings matter: product managers, legal and compliance teams, consultants, and founders and executives. For product managers, the natural workflows are capturing decisions and action items from sprint retros, PRD updates, and stakeholder readouts without writing them up by hand. For sales and client-facing conversations, the Claude integration supports questions like what was decided about pricing across a set of customer calls, returning the agreed terms and the unresolved follow-ups together with the meetings they came from. Legal and compliance users benefit from the fact that audio never leaves the Mac and that transcripts are speaker-attributed. Consultants and executives can use the app to keep a private record of client and internal meetings without sending recordings to a third-party cloud. Speechmark is a one-time purchase that covers up to 3 of your Macs, with no subscription. You can download and try it free. The $49 founding-customer price applies to the first 200 customers; after that the price is $79. It requires macOS 14.2 (Sonoma) or later on an Apple silicon Mac. The download is version 1.0.0 and is 13 MB. The app is distributed as a direct download for Mac, and the Claude Desktop integration installs as a Claude extension rather than requiring a separate API key or configuration. Speechmark's proposition is narrow and clear: record, transcribe, and summarize meetings on your Mac, keep the audio local, skip the bot and the account, and pay once. Its three-step workflow — capture from the menu bar, listen with automatic speaker separation, and write an editorial note using the model of your choice — is what it does, and the Claude connector extends those private notes into a memory your own AI can query. For anyone who treats meeting records as sensitive and wants usable notes rather than raw bullet points, that combination is the whole value.