This project provides an open-source, AI-powered meeting copilot designed for users who need real-time transcription, echo cancellation, and AI assistance during calls. It offers a local-first solution, enhancing productivity and privacy by processing sensitive meeting data directly on the user's desktop, with benefits for researchers and power users seeking to leverage AI for meeting intelligence.

How It Works

The copilot captures both system audio and microphone input simultaneously using native macOS (ScreenCaptureKit/CoreAudio) and Windows (WASAPI) APIs. It then processes these streams through a GStreamer pipeline incorporating the WebRTC AEC3 engine for robust echo cancellation, ensuring clear audio capture. Transcripts are generated in real-time via Deepgram Nova-3 over parallel WebSocket connections. For AI assistance, it integrates with Anthropic Claude or OpenAI, providing context-aware responses and supporting Retrieval-Augmented Generation (RAG) with local documents embedded using @xenova/transformers. All data, including API keys, is stored locally using SQLite.

Quick Start & Requirements

Installation requires cloning the repository and executing npm install, followed by building native audio capture and echo cancellation modules. This process is platform-specific:

• macOS: Requires Xcode Command Line Tools, Node.js (v22+), GStreamer (via Homebrew), and Swift (5.9+).
• Windows: Requires Visual Studio Build Tools (C++ workload), Node.js (v22+), Python (3.12+), Rust, GStreamer (MSVC installer), and CMake.

Users must provide API keys for Deepgram (transcription) and either Anthropic or OpenAI (AI assistance). The setup is non-trivial due to native compilation steps.

Highlighted Details

• Dual-stream audio capture (system audio + microphone).
• WebRTC AEC3 echo cancellation implemented via GStreamer.
• Real-time transcription powered by Deepgram Nova-3.
• Local RAG capabilities for AI context using @xenova/transformers.
• A stealth overlay mode that is invisible to screen-sharing applications.
• Local-first architecture ensuring API keys and data remain on the user's machine.

Maintenance & Community

The project is described as being in "active development" and welcomes contributions through issues and pull requests. Specific community channels like Discord or Slack are not mentioned in the README.

Licensing & Compatibility

The project is released under the MIT license, which is permissive and allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

Linux is explicitly listed as not yet supported. The setup process is complex, involving multiple native compilation steps and dependency management that may require significant troubleshooting, particularly concerning build tools like node-gyp, cmake-js, Swift, and Rust. The open-source version focuses on core functionality; advanced features like billing and sync are part of a separate, paid backend.