Early Alpha
An interface for the swarm of agents working for you.
We're starting with one piece of that: a spatial HUD for smart glasses that lets you see what your AI coding agent is doing and steer it — approve, redirect, answer a question — without touching a keyboard. It's the first build in a broader bet: as more of what you do gets handed to AI agents, on your machine or in the cloud, you'll need an ambient way to see and steer all of them, over open protocols like MCP and A2A rather than one vendor's walled garden.
Approve or redirect a running agent
A simplified version of the real HUD in xr-core. Pick a pending action below, then approve, always-allow, or deny it — the same choice you'd make wearing glasses.
1. Pending Agent Actions
Pick a pending action below to see the permission prompt xr-core would show you.
What I'm building
Three pieces of the xr-core client, in progress.
Bridge Daemon
A lightweight daemon that connects your coding agent to whatever device you're wearing. It can run anywhere on the network, not just your workstation — and longer term, the goal is to talk to agents directly over open standards like A2A and remove the daemon step entirely.
Spatial Intent Parsing
Turning what you say and look at into structured instructions an agent can act on — speech transcription, intent parsing, and the translation layer in between.
Device-Agnostic Client
Built to run across any Android XR device class — headset, wired glasses, or audio-only — instead of locked to one hardware vendor.
Where this stands today
xr-core is early alpha and under active development. The current focus: a native, open workstation daemon and support that isn't locked to one piece of hardware.