Apps — what actually runs on the lab
73
apps published to the web
0
ports open to the internet
66
called AI through the gateway, last 30 days
10,635
AI calls, last 30 days
From the AI gateway's last 30 days.
See what they ask forCall mix
10,635 calls in 30 days, by kind.
Chat carries the load. One app, Dmotion, accounts for about 7,400 of the 10,635.
Apps by machine
Five machines. Every app has a home.
Dev server
AI animation and explainer-video studio. The busiest app on the gateway: about 7,400 of 10,635 calls.
Prompt to animated websites, decks and infographics. This site was made with it.
The lab's monitor.
Voice
Voice agent.
Multi-model agent orchestration.
Browser agent.
Network analysis.
Workflows.
Image & video
AI video front-end on LTX.
Claude-powered animation studio.
Model host
Private chat with an off-the-record mode that is never logged.
Control plane
Skills library.
How it's wired
Same machine
App
its own key
models · budget
Answered
Relay
on its machine
Gateway
routes the call
Switched off
Model / provider
lab or cloud
Gateway first.
Gateway down. Still answered.
Every app gets its own gateway key, with its own model list and its own budget. Cost and usage are booked per app.
The app calls the relay on its own machine, not the gateway directly.
The relay sends the call to the gateway, which routes it to a lab model or a cloud provider.
If the gateway doesn't answer, the relay replays the request straight to the provider with its own backup key. The app's key never leaves the gateway path.
Gateway switched off for 69 seconds: 156 of 156 calls answered.
58 of them by the backup path. Every relay switched back by itself.
10,635 calls in 30 days, and each one booked to the app that made it.