On your Mac
AgentRoom Mac
Registers your folders, starts your agents and keeps every session. Your code and logins stay here, and secrets stay in the Keychain.
AgentRoom Mac app available, Vision Pro in beta
AgentRoom drives coding-agent sessions on your own Mac from a Vision Pro. Your code, your logins and your agents stay where they are.
AgentRoom Mac runs your agents. AgentRoom visionOS is where you work with them.
On your Mac
Registers your folders, starts your agents and keeps every session. Your code and logins stay here, and secrets stay in the Keychain.
Between them, on the same local network
On your Vision Pro
Opens threads, sends turns, edits files and reviews Git changes in the room around you. It runs no agents and holds no credentials.
A useful agent needs your codebase, logins, Keychain and build tools, and they already live on your Mac. Shipping them to someone else's machine for a nicer client is a bad trade.
The headset removes the other limit. A monitor shows two or three files at once. In the room, the conversation, diff, file tree and plan each get a place.
The cost is a Mac that stays awake on the headset's network.
Reasoning, tool activity, diffs and token usage. Restart anything and the conversation is still there.
Manage, search and edit files in a bundled Monaco editor, with language services you can switch on.
Status, staging, commit, pull, push and a spatial history graph. Fixed operations, never command strings.
SVG, Mermaid and editable spatial diagrams, saved as ordinary files. USDZ models open in Quick Look.
AgentRoom drives the agents you already use.
Uses the login already on your Mac
Point it at the executable
Sign in. Turns bill your Cursor plan
Supply the runtime and credentials
Agents speaking Agent Client Protocol can be added too. That path ships switched off.
Three ideas usually discussed apart. Together they change where work happens, who does it and who answers for it.
Computing leaves the flat screen. Work takes up the space around you, so more of a problem stays in view and you move through it instead of scrolling past it.
Software now takes actions for you. Agents plan, write and run the work, so a person's job moves from typing every step to deciding what should happen.
The discipline that holds the other two to account. It settles what a system may touch, how a person checks its work and what happens when it fails, before anything ships.
Each idea is familiar on its own. The paradigm is what happens when they meet. Agents get room to work, and the person keeps sight of all of it and the final say.
Spatial Agentic Engineering keeps a person in charge of every agent in the room.
The Consilience Project argues that technology is not values neutral. Most tools now come from what it calls nihilistic design, which leaves a product's effect on how people think and behave out of the brief. That gets harder to ignore once AI sits in the room around you and competes for your attention.
Spatial Agentic Engineering starts from the other end. The agents do the work, and you decide what they can reach, what they run and what they keep. AI in augmented reality is there to serve the person wearing the headset, so the design decisions carry the ethics.
Read the Consilience Project essayOpen beta. The Vision Pro app is on TestFlight and the Mac app is available now. Expect rough edges.