A local AI assistant for meetings and voice
Wavren AI runs the AI on your computer: Whisper models for speech, Qwen models for summaries, optional Ollama integration for models you already run — local-first, cloud-optional.
The local stack
Speech-to-text runs on Whisper models with GPU acceleration (Metal on Apple Silicon; CUDA and Vulkan on Windows). Summaries and text refinement run on Qwen models through a bundled local engine. If you use Ollama, Wavren AI connects to it directly; if you prefer a cloud model for summaries, bring your own API key — your audio still never leaves the machine.
Why local AI wins for this job
Meetings and dictation generate a continuous stream of sensitive audio. Sending that stream to a cloud service means recurring cost, latency, and a permanent privacy question. Running locally means zero per-minute cost, instant availability, offline operation, and an answer to the privacy question that doesn't require trusting anyone.
Practical hardware, real performance
Wavren AI targets ordinary modern computers — Apple Silicon Macs and recent Windows machines — and picks models to fit your RAM and GPU. 16 GB of memory comfortably runs the best local models.