Meetily — Privacy-First AI Meeting Assistant with Local Transcription

Introduction

Meetily is an open-source AI meeting assistant that runs entirely on your machine. It captures audio, transcribes in real-time using Whisper or Parakeet models, identifies speakers via diarization, and generates meeting summaries through a local Ollama LLM. No audio ever leaves your device.

What Meetily Does

• Transcribes meeting audio in real-time with sub-second latency using local Whisper or Parakeet models
• Identifies individual speakers through automatic speaker diarization
• Generates structured meeting summaries, action items, and key decisions via local LLMs
• Stores all transcripts and summaries locally in an encrypted database
• Works with any audio source including system audio capture and microphone input

Architecture Overview

Meetily uses a Rust-based audio capture engine for low-latency system audio recording. The transcription pipeline runs Whisper.cpp or NVIDIA Parakeet locally, feeding segments into a diarization module that clusters speakers by voice embeddings. A Python backend orchestrates the pipeline and serves the web UI. Summarization uses Ollama to run local LLMs like Llama or Mistral against the full transcript.

Self-Hosting & Configuration

• Install via Docker Compose with GPU passthrough for faster transcription on NVIDIA hardware
• Configure the audio source in settings: system audio loopback, microphone, or both
• Set the Ollama endpoint URL if running on a separate host; defaults to localhost:11434
• Choose your transcription model: whisper-small for speed or whisper-large-v3 for accuracy
• Transcripts are stored in ~/.meetily/data; back up this directory for persistence

Key Features

• Fully local processing with zero data sent to external servers
• Real-time transcription at 4x faster than standard Whisper using optimized Rust audio pipeline
• Cross-platform support for macOS and Windows with native audio capture
• Speaker-labeled transcripts with automatic name assignment after initial identification
• Export to Markdown, JSON, or plain text for integration with note-taking tools

Comparison with Similar Tools

• Otter.ai — Cloud-based with subscription fees; Meetily is free, local, and private
• Granola — Mac-only with cloud summarization; Meetily supports Windows and runs fully offline
• Krisp — Focused on noise cancellation; Meetily provides full transcription and summarization
• Whisper (standalone) — Raw transcription only; Meetily adds diarization, summarization, and a polished UI
• Tactiq — Browser extension for Google Meet; Meetily captures any audio source system-wide

FAQ

Q: Does Meetily require a GPU? A: No, but a GPU significantly speeds up transcription. CPU-only mode works with smaller Whisper models at slightly higher latency.

Q: Which meeting platforms does it work with? A: Meetily captures system audio, so it works with Zoom, Google Meet, Microsoft Teams, Slack Huddles, or any application that outputs audio on your machine.

Q: Can I use a custom LLM for summarization? A: Yes. Any model available through Ollama can be selected in settings, including fine-tuned models optimized for meeting notes.

Q: How much disk space do transcripts use? A: Text transcripts are lightweight, typically under 100 KB per hour of meetings. Audio recordings, if enabled, require about 50 MB per hour.

Sources

• https://github.com/Zackriya-Solutions/meetily
• https://meetily.ai