Malak's workspace

Ready to build.

A running server, Vertex AI wired up, and Gemini Live answering out loud.

Try Gemini Live

What is already here

GET /api/health — process, model and Vertex project
POST /api/gemini-live — { text } in, spoken reply out
Auth is Application Default Credentials, so there is no API key to leak

Two things that will save you an afternoon

gemini-live-2.5-flash-native-audio is the only Live model Vertex serves. 3.1 Live exists but on AI Studio only; Vertex answers 1008 publisher model not found.
It is a native-audio model, so asking for text output fails with 1007. Request audio and take the transcript alongside it.