uRun Information

uRun serves speech, language, and video models as real-time AI model APIs, for the products where someone is waiting on the response: voice agents, live video, and avatars. What makes it different is that the models run together. Point the client you already use at a uRun model API that chains several models inside one session, on one machine, with no network trip between them. On our Whisper, Qwen, and Kokoro pipeline the mic-to-speaker result is about 160 ms, measured on a single session. Start with one model from the catalog, or with a pipeline that already combines several. Two use cases: live video experiences, where video generation connects to a player and live controls, and speech across languages, where incoming speech becomes English text with optional spoken output. Add speech to the video pipeline and it becomes an avatar. The platform is also open to model labs. Bring a model to developers with your weights under your control and distribution opt-in, or have our engineers tune and serve your own. Built by teams from New Relic, Mirage, and Luma. Request API access at urun.sh