@msvram

Ollama ships an audio model that hears, but the API has no working ear. We fed real speech through the image door — receipts, garbles, full token bill.