Audio Summary with Local LLM (Apple Silicon / MLX)
A command-line tool that transcribes and summarizes local audio or video files or YouTube content entirely on Apple Silicon using MLX, picking Voxtral or Whisper for transcription and a Qwen3 model for summarization based on the Mac's available memory.
This tool produces a quick summary of audio and video files, from a local file or directly from YouTube, running entirely on Apple Silicon through Apple's MLX framework, with no NVIDIA or CUDA stack, no PyTorch CUDA stack and no external server. Speech-to-text runs on Mistral Voxtral through mlx-audio, falling back to mlx-whisper on memory-constrained Macs, and summarization runs on Qwen3 through mlx-lm.
At startup it detects the machine's unified memory and automatically selects the highest-tier speech and summarization models that fit, printing the detected memory and chosen models before starting work; models download from Hugging Face on first use and are cached locally. It also supports YouTube downloads, local file summarization, and a transcript-only mode that skips summarization.
Built with Python, requiring an Apple Silicon Mac (M1 or newer), Python 3.12, ffmpeg and uv for package management.