Speechmark

Product Introduction

1. Definition: Speechmark is a native macOS application (category: on-device AI meeting transcription and summarization software) that functions as a private meeting assistant. It captures, transcribes, and generates structured notes from meetings entirely locally on an Apple silicon Mac.
2. Core Value Proposition: It exists to provide professionals with automated, high-quality meeting notes while guaranteeing complete data privacy. Its primary value is enabling users to capture decisions and action items without ever sending sensitive audio data to the cloud, using a one-time purchase model instead of a subscription.

Main Features

1. On-Device Recording & Audio Routing: The app resides in the macOS menu bar. With one click, it captures both microphone input (the user) and system audio (the meeting application like Zoom or Teams) simultaneously. It handles the complex audio routing internally, eliminating the need for a virtual microphone or a bot participant to join the call.
2. Local Speaker-Attributed Transcription: Speechmark processes the audio locally using on-device speech recognition (leveraging Apple's Neural Engine on Apple silicon). It performs automatic speaker diarization (speaker separation), allowing users to label speakers once. Subsequent transcripts are generated as a clean, speaker-attributed script, identifying "who said what" without cloud processing.
3. Flexible, Private AI Summarization: The app generates concise summaries highlighting decisions, action items (with assignees), and a narrative recap. Users can choose their AI model based on privacy needs: fully local options (Apple Intelligence/Foundation Models or a user-run Ollama model) or cloud models (using the user's own OpenAI or Anthropic API key, where only transcript text is sent).
4. Claude Desktop Integration via MCP Server: A unique feature is its one-click integration as a Model Context Protocol (MCP) server for Claude Desktop. This creates a local connector that allows Claude AI to reason over the user's private meeting notes stored on their Mac, answering questions with citations, without the notes ever leaving the device.

Problems Solved

1. Pain Point: The privacy and security risk of uploading confidential meeting audio to third-party cloud servers (e.g., Otter.ai, Fireflies.ai). Many professionals in regulated industries or handling sensitive IP cannot use standard cloud transcription services.
2. Pain Point: The distraction and inefficiency of manual note-taking, which pulls focus away from active participation in meetings and often results in inconsistent or incomplete records of decisions and action items.
3. Target Audience: Product Managers & Project Leads who need accurate records of sprint retros and stakeholder decisions; Consultants & Lawyers who require verbatim, attributable records of client discussions under confidentiality; Founders & Executives in early-stage companies who handle sensitive strategy talks; Compliance Officers who need audit trails without external data processors.
4. Use Cases: Capturing precise action item ownership during project kickoff calls; creating searchable, attributable records of legal or client consultations; quickly generating shareable recaps from lengthy stakeholder syncs; retrospectively querying a private archive of past meeting decisions using Claude AI.

Unique Advantages

1. Differentiation vs. Cloud Competitors (Otter.ai, Fireflies): Speechmark's architecture is fundamentally opposite. Competitors are cloud-first: audio is uploaded, processed on their servers, and stored in their cloud, requiring subscriptions and accounts. Speechmark is device-first: all core processing is local, requires no account, and uses a one-time fee.
2. Differentiation vs. Other On-Device Tools (Granola): While some competitors also offer on-device transcription, Speechmark distinguishes itself by refusing to sync notes to its own cloud by default and by not using anonymized user data for model training. Its business model is purely software sale, not data-informed. The deep, local Claude Desktop integration is also a unique capability enabled by this fully local data stance.
3. Key Innovation: The combination of a fully local, menu-bar-based audio capture engine with a flexible "bring-your-own-model" AI summarization layer. This gives users unprecedented control over the privacy-performance trade-off, from fully offline operation to using the most powerful cloud models without ever exposing the raw audio.

Frequently Asked Questions (FAQ)

1. Does Speechmark work without an internet connection? Yes, for core functionality. Recording, local transcription (using Apple's on-device model), and summarization (using Apple Intelligence or a local Ollama model) work completely offline. Internet is only required if you choose to use OpenAI or Anthropic's cloud models for summarization.
2. How does Speechmark record both my voice and the meeting audio without a bot? The app uses macOS system-level audio APIs to capture your microphone input and the computer's system output (the audio from your meeting app) simultaneously. It mixes these audio streams locally on your Mac. No virtual participant or bot joins the call; it's akin to recording the audio playing through your speakers and your voice through the mic.
3. What are the specific system requirements for Speechmark? Speechmark requires macOS 14.2 (Sonoma) or later and an Apple silicon Mac (M1, M2, M3, or later chips). This is mandatory because it relies on the Neural Engine for efficient, on-device speech recognition and transcription.
4. Can I export my Speechmark meeting notes and transcripts? Yes. Speechmark allows you to export your meeting data, including transcripts and summaries, in portable formats like Markdown and JSON, as well as directly into a SQLite database for advanced querying or integration with other local tools.
5. Is the "Claude Desktop" integration secure and private? Absolutely. The integration works by installing a local MCP (Model Context Protocol) server that Claude Desktop can access. When you ask Claude a question about your meetings, Claude Desktop reads the relevant note files directly from your local disk. No meeting data is sent to Anthropic's servers via this connector; it remains entirely on your Mac.