Íris — One Brain, Multiple Bodies

Today's AI assistants are trapped inside a single app on a single device. Íris inverts that model: every user gets one persistent AI Brain, and their devices become interchangeable "bodies" it acts through. Sign up and your own isolated Brain (built on the Hermes agent) is auto-provisioned in the cloud — no setup, no config files. Each device then connects over a lightweight WebSocket capability protocol and announces what it can do: a desktop (Electron), an Android phone, and a Wear OS watch each advertise capabilities like speech, local LLM, GPS location and push notifications. The Brain routes every action to whichever body can serve it — pulling a live GPS fix from your phone, speaking a response through your desktop's GPU-cloned voice, or firing a notification to your watch. Inference is AMD end-to-end. Gemma runs across three tiers chosen live by the Iris Orchestrator: locally on a Radeon RX 9060 XT via LM Studio, on an AMD Instinct MI300X via vLLM, and a cloud fallback. A kill-switch failover means that if your local GPU drops mid-conversation, the AMD cloud takes over seamlessly and the model indicator updates in real time. Custom voice comes from a local F5-TTS clone running on the same GPU. The result isn't a chatbot in a window — it's one mind that follows you across every screen and sensor you own. One Brain. Multiple Bodies. Accelerated by AMD.