Plain questions, direct answers
Is anything monitoring me? Is anything hiding from the process list? What starts on its own? Each audit answers one question.
Is my machine still mine?
A guardian for machines that run AI agents.
Agents run code with your credentials. JARVIS watches the machine they run on: what is running, what could take or keep control, and what has quietly changed. It works locally, keeps evidence, and tells you what it cannot see.
The Lattice. A sweep passes over the machine, an intruder is flagged and held, then released on undo. One panel is left open on purpose: the blind spot it admits to.
How it works
Processes, autostarts, network activity, sensors and firewall state across Windows and Linux views.
Events are joined into a timeline so a change has a before and an after.
Findings come with the evidence behind them and with what could not be seen.
Reversible containment only when you have switched it on. Anything more waits for you.
What it does
Is anything monitoring me? Is anything hiding from the process list? What starts on its own? Each audit answers one question.
It shows what runs because you opened an editor, a repository or a shell, which is exactly where an agent works.
What is observed is recorded before anything responds to it. Something that is merely absent from a later scan is logged as not observed, never as fixed.
Watch, ask, propose, act. Automatic actions are limited to reversible containment, every action is in a ledger, and there is an undo.
Public threat feeds are vetted by a deliberately simple checker. Nothing fetched is ever executed, and intelligence can never clear a finding.
JARVIS compares current attacker techniques with what its own sensors cover and reports the gaps as a work list.
The dashboard only listens on your own machine. There is no telemetry by default, and the core depends on almost nothing but the language it is written in.
Illustration
In practice
jarvis spyjarvis hiddenjarvis toolchainjarvis secretsjarvis evasionjarvis gapsWhat it is not