System vitals
SimulatedCPU
GPU
SYSTEM RAM
GPU MEMORY
Compute activity
6 CPU coresRun a sample request to watch the simulated workload shift.
Service stack
Sample stateCommand console
Ready for inputScripted demo: try status, find a file, code, image generation, or a theme change.
Try a command
Choose a scenarioModel routing
AutomaticA request is classified, then assigned a tool or model. Larger workloads compete for the same GPU memory.
About this build
Based on Matthew’s JarvisOS repository: an Ubuntu home lab with a holographic HUD, local inference, file tools, and a voice interface.
Read the project notes ↗A workspace you can talk to.
Search and open these fictional sample documents. In JarvisOS, file tools search names, folders, and document text; this demo uses an in-memory sample collection.
Sample workspace
3 filesDocument preview
Sample dataSelect a file to inspect its contents.
Follow the work.
A simulated execution log showing request classification, model selection, and tool events. This is a demonstration of system operations, not a model’s private reasoning.
Execution timeline
Resource-aware routing
The real project also supports cloud models and shared spending limits. This public demo never calls those services. Model labels below illustrate the original Qwen3 / GPT-OSS configuration; the live project’s configuration can evolve.
Your interface. Your preferences.
Changes affect only this demo tab. Try a different hologram color or select a model route.
Appearance
Also respects your device’s reduced-motion preference.
Uses your browser’s speech voice when available, not Jarvis’s Piper engine.
Demo controls
Simulates model selection; no models run in this page.
Reset the demo
Clears messages and execution history, and restores default settings.