tome-app  by Gremble-io

Local meeting transcription and capture for Obsidian AI agent pipelines

Created 2 weeks ago

New!

359 stars

Top 78.1% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

Tome is a macOS application designed for local meeting transcription and seamless integration into Obsidian vaults, targeting users who prioritize data privacy and automated AI agent workflows. It addresses the gap in existing tools by offering an end-to-end, on-device solution that captures audio, transcribes it, identifies speakers, and outputs structured Markdown files directly into a user's Obsidian vault, eliminating cloud dependencies and API key requirements.

How It Works

Tome leverages Apple Silicon's capabilities for local, on-device processing. It captures audio via microphone or system audio, specifically filtering input from supported conferencing applications (Zoom, Teams, Slack). Transcription is handled by Parakeet-TDT v3, supporting 25 European languages with auto-detection, while speaker diarization is performed post-session using pyannote. The core innovation lies in its output: plain Markdown files (.md) containing structured YAML frontmatter (including type, timestamp, attendees, and tags), designed to be immediately consumable by downstream AI agents within an Obsidian environment. This approach ensures data remains entirely local, enhancing privacy and enabling automated knowledge base management.

Quick Start & Requirements

  • Primary Install/Run: Build and install via the provided script: ./scripts/build_swift_app.sh. The application installs to /Applications.
  • Prerequisites: Requires an Apple Silicon Mac, macOS 26+, and Xcode 26.3+.
  • Initial Setup: First launch requires downloading the Parakeet ASR model (~600MB), which is cached thereafter. Users may encounter macOS Gatekeeper security prompts for unsigned applications; a right-click "Open" action or building from source bypasses this.
  • Documentation: README serves as primary documentation.

Highlighted Details

  • On-Device Processing: Full transcription and speaker diarization occur locally, ensuring no data leaves the user's machine.
  • Vault-Native Output: Generates .md files with YAML frontmatter (type, created, attendees, tags, source_app), optimized for Obsidian and AI agent pipelines.
  • Intelligent Audio Capture: Filters system audio to capture only the selected conferencing application, excluding notifications or other background sounds.
  • Privacy-Focused: No audio is saved to disk; only text transcripts are retained. The app window is hidden from screen sharing by default.
  • Multilingual Support: Parakeet-TDT v3 supports transcription in 25 European languages with auto-detection.

Maintenance & Community

The provided README does not detail specific community channels (e.g., Discord, Slack), active maintainers beyond the initial author, or a public roadmap. Sponsorships or partnerships are not mentioned.

Licensing & Compatibility

  • License: MIT License.
  • Compatibility: Permissive for commercial use and integration into closed-source applications due to the MIT license. However, the application itself is strictly limited to Apple Silicon Macs running macOS 26+.

Limitations & Caveats

Tome is exclusively compatible with Apple Silicon Macs and requires macOS 26+. Speaker diarization is noted as imperfect and runs post-session, meaning remote speakers are presented as a single stream during live calls. The application may require manual bypassing of macOS Gatekeeper until a signed release is available.

Health Check
Last Commit

6 days ago

Responsiveness

Inactive

Pull Requests (30d)
16
Issues (30d)
9
Star History
360 stars in the last 15 days

Explore Similar Projects

Feedback? Help us improve.