Conversationaly: A Privacy-First AI Meeting Assistant

Conversationaly is an innovative desktop application designed for macOS, Windows, and Linux that revolutionizes how meetings are recorded and summarized. This privacy-centric tool captures both microphone and system audio, transcribing live dialogues and generating concise summaries entirely on your device, ensuring no data is sent to the cloud unless explicitly configured by the user.

Key Features

• Local Transcription: Utilizes approximately 87 models from 16 different families (including Whisper and Parakeet) for real-time transcription without relying on external servers.
• Live Transcripts: Stream native transcription models provide continuous transcription while speaking, while batch-only models operate using voice activity detection.
• Built-in AI Summaries: Integrates a llama-helper sidecar that runs Gemma 4 locally for generating summaries, which can also transcribe audio as an LLM.
• Custom LLM Integration: Supports varied LLMs for summary creation such as Ollama, Claude, Groq, and OpenAI, enabling flexibility based on user preference.
• Optional Cloud Integration: Users can opt for cloud-hosted STT services, like Deepgram or ElevenLabs, should they prefer remote processing.
• Audio Mixing Capabilities: Merges microphone and system audio with RMS-based ducking and clipping prevention for enhanced clarity.
• Import & Enhance Feature: Provides functionality to transcribe existing audio files and re-transcribe past meetings with different models or languages.
• Summary Template Selection: Users can select pre-defined structures for summaries and set distinct languages for both summaries and spoken content.
• GPU Acceleration Support: Compatible with Metal on Apple Silicon, CUDA for NVIDIA, and support for Vulkan and ROCm on AMD.
• Local Storage: Transcripts, meetings, and audio models are securely stored in a SQLite database on the user’s machine.

How It Works

Upon first launch, Conversationaly downloads essential models for transcription and summarization. Subsequent meetings involve:

1. Capturing combined mic and system audio, which is recorded.
2. Resampling this audio for use in real-time transcription.
3. Providing summaries by sending transcripts to the configured LLM provider (local by default).

The entire process is executed locally to maintain user privacy, with cloud integrations being optional.

Development and Contributions

Developers interested in contributing to Conversationaly can run a local development environment using PNPM and Tauri commands. The project encourages collaboration through issues and pull requests, with detailed guidelines available in the contributing documentation.

Conversationaly stands as an essential tool for professionals seeking a secure and efficient method to manage meeting documentation without compromising on privacy.