
Gemini workspace with live voice

Gemini workspace with live voice
THINK is a multimodal AI workspace built by Katlego Mokoto (Mens Mercatus). One React 19 / TypeScript app for chat, image, live voice over WebSocket, Net Sites, Net Code and a learning path, powered by Gemini. How it is built: the browser never holds the API key. Every model call goes through a same-origin proxy to a C# / ASP.NET Core 8 minimal API, running as a systemd service on an Ubuntu VPS behind nginx. The API uses an OpenAI-compatible client, so the model provider can be swapped in config without a UI rewrite. Production habits from the first public version: - Exponential backoff on HTTP 429 and 5xx errors - Daily per-IP caps on chat, image, video and live voice. If you hit one, that is the control working, not a bug - A labelled demo mode: if the health check fails, the screen says it is in demo, instead of pretending to be live Limits: THINK is a portfolio workspace with caps, not an unlimited production assistant. The feed content inside THINK is product demo content, not a claim about real users. Try it: open Chat and ask one question, then try Live Voice. The case study at mensmercatus.online/case-studies/assist/ covers the architecture. Mens Mercatus builds AI agents and LLM automation for SMEs up to enterprise, for clients globally, with a human yes before any send, payment or other irreversible step. Website: https://mensmercatus.online Contact: [email protected]