---
title: "Local cybersecurity — audits, fixes and CRA compliance | AgentVUE"
description: "AgentVUE inventories your machines and repositories, applies fixes with your approval and writes a dated report. php2shell, CISA alerts, CRA obligations: act before the incident. From €49/month."
canonical: "https://agentvue.ai/cybersecurity/"
language: "en"
---
AgentVUE cybersecurity

# Your machines, checked and fixed. Locally.

Inventory, real advisories, fixes applied with your approval, and a dated report. Nothing leaves your machine — and the deadlines do not wait.

[Get AgentVUE→](/pricing/)[Meet the cyber analyst](/ai-employees/cybersecurity-analyst/)

-   Runs on Windows & Linux
-   Your models, your keys
-   No account. No telemetry.
-   From €49/month

Cyber · Local passLocal

This run**Machine and repositories**

-   Inventory**142 packages**
-   Advisory match**3 findings**
-   Fix**1 prepared**
-   Report**Dated · local**

InventoryMatchFixReport

Critical**1**High**2**Local**0 B**

Nothing changes without your approval.

Advisories

## What is circulating.

The flaws worth knowing, with their dates and sources.

AllPHPNode.jsCISA KEV

PHPCritical

### php2shell

Two unauthenticated remote code execution flaws (CVE-2026-109618 and CVE-2026-109619), fixed in PHP 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35. CISA published an alert on 21 October 2026 urging immediate patching, and Cybermalveillance.gouv.fr relayed an AlerteCyber on 22 October.

[CISA alert](https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching)

Node.jsHigh

### October releases

Security releases shipped on 21 October 2026 (26.12.1, 24.22.1, 22.23.4), including a Brotli decoder out-of-bounds write enabling code execution (CVE-2026-111497) and two TLS certificate hostname bypasses.

[Node.js security releases](https://nodejs.org/en/blog/vulnerability/october-2026-security-releases)

CISA KEVHigh

### Six new entries

Six new entries were added on 20 October 2026 (including four Plex CVEs) and the #StopRansomware SafePay advisory (AA26-293A) was published the same day.

[CISA advisories](https://www.cisa.gov/news-events/cybersecurity-advisories)

The CRA clock

## The deadline is already running.

Part of the Cyber Resilience Act already applies.

10 Dec 2024**In force**11 Sep 2026**Reporting**11 Dec 2027**Main obligations**

### The regulation applies.

Regulation EU 2024/2847 has been in force since 10 December 2024. Software vendors of 5 to 50 people are manufacturers under the regulation, and their products with digital elements are in scope.

### Incident reporting is live.

Actively-exploited incident reporting applies: early warning within 24 hours, full notification within 72 hours and a final report within 14 days, through ENISA’s single reporting platform.

### The main obligations arrive.

The main obligations arrive on 11 December 2027. The inventory, the list of exposed versions and the fixes applied with your approval are what the technical file asks for.

[European Commission — Cyber Resilience Act](https://digital-strategy.ec.europa.eu/en/policies/cyber-resilience-act)[European Commission — CRA incident reporting](https://digital-strategy.ec.europa.eu/en/policies/cyber-resilience-act-reporting)

One pass

## One pass. Four steps.

Inventory, match, fix, report. All on your machine.

One local passPlay the loop

01**Inventory**02**Match**03**Fix**04**Report**

### Inventory

Packages, services, ports, users and scheduled tasks, read from the machine.

### Match

Installed versions are compared against npm, PyPI, vendor and CISA advisories.

### Fix

The fix is proposed, snapshotted and applied with your approval.

### Report

A dated local report records what changed and what is pending.

What it replaces

## A PDF that goes stale, or a machine that stays checked.

An external audit is a snapshot. The local pass runs again.

01

### External audit

Thousands for a point-in-time review. A PDF lands, then the week after it is already out of date.

02

### AgentVUE

The machine is checked, fixed and documented on every run. From €49 to €199 per month, depending on the plan.

![AgentVUE cybersecurity report showing security posture, 64 findings across 36 advisories and the remediation summary.](/images/product/cybersecurity.webp)

**Cybersecurity — dated report**The local security posture report with findings, exposure and the recommended fixes.

Built with AgentVUE

## The workstation checks itself.

AgentVUE runs its own security check on every release. You are buying the machine that has already done the work.

[See the showcase](/showcase/)

![AgentVUE release check showing a completed security pass, 64 findings across 36 advisories and the dated report.](/images/product/cybersecurity.webp)

**Cybersecurity — AgentVUE’s own release check**The security posture report AgentVUE produces from its own release check.

The roles

## Two cyber roles. One machine.

The analyst reads. The engineer fixes. Both run on the same licence.

[

### AI cybersecurity analyst

Reads the exposure, ranks the findings and explains what matters first.

Read-only by design](/ai-employees/cybersecurity-analyst/)[

### AI cybersecurity engineer

Prepares the fix, snapshots the state, applies it with your approval and checks the result.

Approval before every fix](/ai-employees/cybersecurity-engineer/)

Start here

## Start with the first inventory.

Pick your plan, install and run the first pass. The report starts with what is actually installed.

[Get AgentVUE→](/pricing/)[Meet the cyber analyst](/ai-employees/cybersecurity-analyst/)

Questions

## Questions, answered.

The inventory, the fixes, the reports and the deadlines.

How does the licence work?

You buy online and receive a signed licence file. The software runs offline: no activation server, no telemetry. The version you bought keeps working for good; updates are covered while your support window is open.

Do I get the source code?

The workstation is built on open-source components — the Code-OSS editor core, the AgentCode runtime and its MCP servers (MIT or Apache-2.0 depending on the component) — and ships as a local application you can inspect and script. The AgentVUE suite remains licensed software: Ultra grants internal modification, not redistribution.

Which AI models does it use?

Yours. AgentVUE talks to any OpenAI-compatible endpoint — a local model, your own server, or a provider you trust. Keys are stored in SecretStorage, never in a config file.

What can the agents actually do?

What a careful operator does on a computer: write and review code, run tests, drive a browser, generate images, video and music, manage Git issues and pipelines, and keep recurring tasks moving.

Is my code sent somewhere?

No. The work happens on your machine. The only outbound calls are the ones you configure: your AI endpoint, your media endpoints, your voice endpoints.

What about security?

The cyber track inventories your machines and repositories, checks versions against published advisories and proposes the fix. Nothing is applied without a snapshot and your approval.

Can I sell what I create?

Max and Ultra, yes. Max covers commercial use inside your organisation, including selling the outputs you produce. Ultra adds white-label reports, agency/OEM delivery and internal modification of the software.

Is support a phone call?

No. Email or ticket, best effort, five business days. The product is built to be used without meetings.

Where does AgentVUE come from?

It began as one person's tool for running several projects without stitching five apps together. The local rigs came next, then the agents. The founder tells the whole story, including the risks that shaped the product, on benoit.ac.

[Read the origin story on benoit.ac](https://benoit.ac/blog/what-is-the-origin-story-of-agentvue/)

Next step

## Run the first pass.

Pick your plan, install, run the first pass today. The dated report starts with what is actually installed.

[Get AgentVUE→](/pricing/)[Talk to the founder](/consultancy/)
