Software that watches the noise
and tells you what matters.
AROK builds local-first AI for the streams you can't read yourself — your machine's telemetry, the world's headlines. Detection is deterministic. The AI only explains. Nothing leaves your device.
Open source (MIT) · No account · No telemetry · Works offline
Dashboards show you data.
AROK tells you the story.
You are drowning in streams nobody has time to read — process tables, network sockets, drive sensors, a thousand headlines an hour. We split the problem in two: math finds the signal, language explains it. The model never decides what's wrong. It only says it out loud.
Continuous, not on-demand
AROK runs in the background and keeps history, so "unusual" means unusual for you — not unusual against some vendor's average.
Deterministic, never guessed
Thresholds, trends and transparent heuristics decide what's flagged. Every finding can be opened and audited. No hallucinated alerts, ever.
Plain English, on-device
A local model turns findings into a sentence a human can act on. No inference leaves the machine. No prompt becomes someone else's training data.
Same engine. Two streams.
One watches the machine in front of you. The other watches the world around it. Both run on your terms.
AROK Monitor
The system monitor that explains itself. Nine surfaces — processes, network, services, SMART, cleanup, analytics, AI insights, upgrades — narrated by a model running on your own hardware.
“svchost.exe has climbed steadily for 20 minutes and now holds 78% CPU — unusual for this time of day on your machine.”
NEXUS
Your news, your way. A feed ranked by topics you rate — not by what an engagement algorithm decided you'd click. Weather, alerts, and a daily digest in one screen.
The cheapest way to protect data
is to never send it.
Most “AI-powered” monitoring is a pipe from your machine to someone else's. We took the other road, and it turns out the other road is also faster and free.
AROK
- ✓ Metrics, history and inference stay on device
- ✓ Server binds to 127.0.0.1 only — nothing on your LAN
- ✓ Works with the network cable pulled
- ✓ No account, no sign-in, no identity to leak
- ✓ MIT source — audit every line yourself
- ✓ Zero telemetry, zero ads, zero paid tier
Cloud-dependent tooling
- × Your process list uploaded for “analysis”
- × Agent phones home on a schedule you don't control
- × Offline means blind
- × Account required, breach surface created
- × Closed detection logic you must take on faith
- × Usage analytics, upsells, per-seat pricing
Your model, your key, or no model at all.
Pick from a curated menu of open-source GGUFs — Qwen 2.5 1.5B through Mistral 7B — or point AROK at a model already on your disk. Whatever you choose, the app still works.
Local model
Private, offline, free. Downloaded once, then never touches the network again. This is the default and the recommendation.
Anthropic Cloud
Optional. Bring your own API key if you want a larger model for narration. Off unless you turn it on.
Deterministic templates
No AI configured? Findings still render as clear written summaries. Detection never depended on the model in the first place.
Questions, answered.
Yes. Open source under MIT, no paid tiers, no ads, no telemetry. AROK Monitor is funded only by optional, clearly-disclosed Amazon affiliate links inside the Hardware Upgrades tab. If you never open that tab, AROK costs nothing and sells nothing.
No. Metrics, history and AI inference all stay on your machine, and the local server binds to 127.0.0.1 so nothing is reachable from your network. The internet is only touched if you choose to download a model, refresh upgrade pricing, or enable optional cloud narration with your own key.
Localhost Host and Origin checks block DNS rebinding, and per-session tokens block CSRF from malicious pages. Demo mode simulates control actions — kill process, block IP, stop service — instead of executing them, so you can look around safely first.
AROK Monitor runs on Windows 10 and 11 — grab the installer from GitHub Releases and run it; updates download quietly in the background and apply with one click. NEXUS runs in any modern browser with nothing to install.
That's the point. Every finding exposes the heuristic behind it — click Investigate on any flagged network endpoint to see exactly why it scored the way it did, or mark it safe. And the whole codebase is MIT on GitHub if you'd rather read the source than trust the UI.
See what your machine
has been trying to tell you.
Free, open source, and fully offline. Install in under a minute — or just open NEXUS and read something worth reading.
View source on GitHub · MIT License · As an Amazon Associate, AROK earns from qualifying purchases made through links in the Upgrades tab.