Ollama ships local text-to-speech with an OpenAI-compatible endpoint
Version 0.40.11 adds spoken replies on macOS via ollama run, plus d3, a ported decision-model family in four sizes for classification without generating tokens.
Ollama 0.40.11, released on 24 October, adds local text-to-speech on macOS. Running ollama run phonon "Hello from Ollama." speaks the reply aloud as it generates, and the same audio is available to your own app through a new OpenAI-compatible /v1/audio/speech endpoint.
The same release ports d3, vLLM Semantic Router’s decision models, into Ollama. d3 ships in four sizes, 2B, 4B, 9B and 27B, and takes text, structured state or images to return a classification, a yes/no answer or a probability score, rather than generated text.
ollama run d3 "A customer says their card was charged twice." --ask "Is this a billing issue?"Why it matters: both features skip a cloud call for jobs that don’t need a full chat model. A local TTS endpoint with the OpenAI shape drops straight into code written for OpenAI’s API, and a decision model gives a cheap, fast classification step for triage or routing without the latency of a generative response.
The caveat: speech generation is macOS-only in this release, with no word yet on Windows or Linux support.