Your terminal can now talk back to you, and the funny thing is that it does it without calling any cloud API. On Saturday, Ollama shipped 0.40.11 with text-to-speech on macOS. The first speech model is Phonon, a family of small TTS models from Gradium. You pull it with ollama pull phonon and run it like any other model. The bit developers will actually care about is the API. Phonon works with the OpenAI-compatible /v1/audio/speech endpoint on localhost. If your code already uses the OpenAI SDK for speech, the docs say you only need a new base URL. The response streams while audio is being generated, so playback can start before the full text is done. You can even pipe a chat model's reply straight into Phonon, and it starts speaking after the first sentence. Output is 24 kHz mono, as WAV by default or raw 16-bit PCM if you ask for it. The docs list some limits that are easy to trip over. Each request takes up to 4,096 characters, although ollama run splits longer text for you. Prices, symbols, emails and URLs get spelled out, while numeric dates are read as written, so the docs suggest writing dates out in words. Unsupported options like speed or instructions return a 400 error instead of being quietly ignored. The licensing deserves a careful read before you ship anything. The default English phonon model is Apache 2.0. The 40M French, German, Spanish, Portuguese and multilingual variants are CC BY-NC 4.0, which rules out commercial use. Ollama also says that on an M5 Max MacBook Pro it ran 35 percent faster than Gradium's own runtime, which is their benchmark rather than an independent one. Speech is Apple Silicon only for now, with Windows and Linux support listed as coming soon.