When your machine
slows down, know why.

A private instrument for your Linux machine.
Clear signals for you and your personal AI.

uv tool install sensitivebox
See it before the call

Readings stay with you.

No telemetry to third-party cloud or AI

No data collected

No analysis or surveillance

Your own AI

Use AI you already trust.

True ZDR

You can run a mighty local model

0

Make room for what matters.

Your work, your demo and your next call share one machine. Give your AI the context before the call. Let it check what the machine can tell you.

Instrument

Your machine, before the call

The demo is running. Your background model is still working.

Sample readings and actions. Nothing on your machine is being changed.

Watching locally

Example readings. This page has no access to your machine.

AMD EPYC 9004 processor with its heat spreader raised above the chiplets
NVIDIA Blackwell Ultra B300 GPU package
NZXT Kraken Elite 360 RGB with three illuminated radiator fans and its round LCD pump display
Front and ball-grid underside of Samsung LPDDR6 memory
Samsung 9100 PRO 8 TB NVMe SSD
ASUS Pro WS 3000 W Platinum power supply, with its fan grille and power connectors
A single NVIDIA DGX Spark desktop AI computer

The evidence stays
within reach.

The alerts are doing the watching. Staring at a hundred values mid-build is optional misery. Open the dashboard when you want to follow a signal, see why an alert fired, or simply enjoy a clear view of the machine. Then get back to your work.

Explore the documentation
Actual Sensitive Box alerts view in the light theme, captured from the heatwave demo.

A small model. A clear picture.

A tiny local model becomes useful when it has something solid to work with: observations, history and the context you provide. What is making the machine wait? Is it running out of headroom? Would cleaning help more than an upgrade? Give your AI the evidence to reason with.

Sensitive Box
Observations
History
Reasons behind alerts
Your modelChosen by you
Your context. Your limits.

Understand the cause.
Choose a proportionate action.
Check whether it helped.

Only the permissions you give

“We provide the instrument. You bring the AI. Keep it local, choose its skills, and decide what it is allowed to do.”

Several of your boxes.
One private view.

The laptop in front of you, the workstation, the GPU box, the machine at home. Each keeps its own history and keeps watching. You reach the others over SSH, with your own keys.

Nothing in the middle holds the readings. Your AI stays inside that boundary.

This machineYou are here Its own history

Direct SSH · your keys

WorkstationIts own history
GPU boxIts own history
Home serverIts own history

All yours. No hosted relay. No central copy.

No telemetry.

Nothing collected for someone else, or “to improve your experience.”

No central ingestion.

Your boxes talk privately. There is no Sensitive Box server in the middle.

Your own devices.

Built to look after your machines. No surveillance of other people.

Open source.

Community verified, maintained and improved.

Truly free.

No licence, no fees, no freemium, no ads.

Community

Good ideas travel.
Your data doesn’t have to.

Borrow a skill, a dataset, an automation, or an alert pack for your device. Make it your own. Share something useful when you choose.

Explore the project on GitHub

A starting point for your AI

Use our skill or adapt a community skill to the way you work. The model, the context and the permissions remain yours.

A contribution you control

Share an idea, an alert rule or a redacted hardware dataset. You choose what leaves your machine.

How to contribute

Room to grow

Build loops and automations around your instrument as you need them. Start with visibility. Give it more responsibility when you’re ready.

Give yourself a little less to keep in your head.

Sbox observes the tools so you can focus on your work.

uv tool install sensitivebox

Then sbox enable to keep watch. sbox open when you want a closer look.

Read the installation guide

macOS via Homebrew is coming soon. Flatpak follows later.