Binary Rain HUD

Your PC is already talking. Binary Rain HUD turns local Windows telemetry into a living display, then lets GPT-5.6 explain what deserves attention.

Comment

Inspiration

I wanted the screen on my tower to do more than go black after five minutes. Most system monitors are dense engineering tools, while most screensavers are decoration. Binary Rain HUD sits between them: useful while I am working, cinematic when I step away, and honest about what it collects.

What it does

Binary Rain HUD is a local-first Windows telemetry dashboard and idle screensaver. It visualizes CPU, RAM, NVIDIA/AMD/Intel GPU activity, VRAM, local Ollama models, and optional creative services in real time. The binary rain responds to system pressure, so the display itself becomes a readable signal.

Clicking RAM opens a local process drawer. Clicking Axiom requests an on-demand system-health brief. Without an API key, Axiom provides a deterministic local analysis. With an OpenAI key, GPT-5.6 receives a minimized aggregate snapshot and returns a structured severity, explanation, and up to three safe next steps.

Nothing is sent automatically. The server binds to 127.0.0.1, and hostnames, process names, PIDs, files, keystrokes, screen contents, and local model names are excluded from the OpenAI payload.

After the configured idle period, the dashboard becomes a fullscreen ambient display. Mouse movement and clicks remain interactive; Escape exits.

How I built it

  • Dependency-free Node.js HTTP server and Windows telemetry collectors
  • HTML Canvas rendering for binary rain and utilization trends
  • Layered GPU detection using NVIDIA tooling plus vendor-neutral Windows counters for AMD, Intel, integrated, multi-GPU, and no-GPU systems
  • PowerShell and Windows Task Scheduler for the least-privilege idle monitor and fullscreen lifecycle
  • OpenAI Responses API with strict structured output and store: false
  • Automated tests for privacy boundaries, telemetry fallbacks, settings validation, and path safety
  • A reproducible Python, PowerShell, and FFmpeg pipeline for the privacy-safe demo video

How I used Codex and GPT-5.6

Codex was my implementation and review partner. It accelerated the browser interface, localhost server, GPU telemetry fallbacks, Windows idle behavior, foreground debugging, privacy hardening, automated tests, release packaging, and demo-video pipeline. I kept the product direction, consent model, and key technical decisions human-led.

GPT-5.6 powers the optional Axiom brief. The application sends only allowlisted aggregate telemetry, requests schema-constrained output, and disables storage. A local fallback keeps the full interaction testable without a key.

Challenges

The hardest problem was not collecting more telemetry—it was drawing a crisp privacy boundary. Process details are valuable on the local display but unnecessary for an AI health brief, so the server constructs the model payload from an explicit allowlist instead of forwarding browser state.

Windows idle behavior was another surprisingly deep challenge. The final design uses a long-running current-user monitor so resuming mouse activity does not immediately dismiss the display. It stays interactive until the user presses Escape.

What changed during Build Week

The project began as a pre-event binary-rain dashboard and idle-launch prototype. During Build Week I substantially extended it with the GPT-5.6 Axiom brief, structured Responses API output, a privacy-minimized payload, responsive telemetry dock and settings, vendor-neutral GPU collection, Windows foreground behavior, security hardening, automated tests, release packaging, and a code-generated public demo. The dated repository history documents that work.

What I learned

Trust is a product feature. Showing what stays local, making transmission a deliberate click, and minimizing the payload made the AI feature feel more useful—not less. Structured output also kept the interface stable across very different machine states.

What's next

  • Signed Windows installer and tray controller
  • Cross-platform collectors for macOS and Linux
  • Opt-in historical baselines and anomaly detection
  • User-defined service probes and workload profiles
  • Voice health briefs through the OpenAI Realtime API

Run it

On Windows 10 or 11 with Node.js 18+, clone or download the repository and double-click start.bat. The zero-key local experience works immediately. Optional OpenAI setup and screensaver installation are documented in the README.

Built With

Share this project:

Updates