Meetily

Meetily is a desktop application built with Tauri and Rust that runs speech-to-text models directly on local workstation hardware. It captures conference audio and microphone input at the operating system level without dispatching a bot into the call. Audio recordings and transcripts remain on the local disk, while summaries can run through Ollama or a private inference endpoint.

Why Meetily instead of Otter.ai

Otter.ai sends a virtual recording bot into video conferences and streams raw audio to external cloud servers for processing. For engineering organizations handling architecture reviews or incident disclosures, sending real-time speech to a third-party vendor introduces compliance and data leakage concerns. Meetily intercepts system audio output and microphone input directly on the host machine. No bot appears in the attendee list on Zoom, Google Meet, or Microsoft Teams, and audio streams never leave the workstation during transcription.

The operational tradeoff centers on compute and administrative convenience. Otter.ai provides a hosted collaboration layer with centralized search, team workspaces, and automatic bot calendar scheduling for a recurring monthly seat fee. Meetily replaces per-minute fees and external storage by running Whisper or Parakeet models locally, accelerated via Metal, CUDA, or Vulkan. Teams can summarize calls offline with Ollama or route text transcripts through a private OpenAI-compatible endpoint. Organizations gain complete governance over sensitive spoken data, but engineers must manage their own disk capacity, backups, and workstation compute.

What it does well

• Runs Whisper.cpp and ONNX Parakeet speech models locally with hardware acceleration on Apple Silicon, NVIDIA GPUs, and Vulkan-supported platforms.
• Intercepts audio directly from system devices, eliminating the need for an external participant bot to join access-controlled calls.
• Connects to local Ollama servers or private OpenAI-compatible inference APIs to generate summaries without exposing prompts to public cloud providers.
• Exposes a local HTTP API, CLI, and Model Context Protocol server that lets coding assistants like Claude Desktop or Cursor query meeting context directly.
• Imports existing audio files and supports retranscription with different speech models, languages, or prompt templates.
• Exports structured summaries and transcripts into Markdown, Word, and plaintext files that integrate cleanly with Git repositories and internal wikis.