Rysa
An Android app I wrote to audit my own network. It runs nmap in Kali NetHunter on a rooted phone and turns the results into a prioritised report.
I scanned Metasploitable 2 with it, a deliberately vulnerable lab machine. The result is a 79-page report. It is an exercise in my own lab, with no client.
Sample report from Rysa · A scan of a Metasploitable 2 machine
- criticalvsftpd 2.3.4 with a backdoor (CVE-2011-2523, CVSS 10.0) and anonymous access.
- criticalAn open administrator shell on port 1524 with no authentication.
- criticalMySQL 5.0.51a with an empty root password; the whole file system exported over NFS.
- 30findings
- 5critical
- 7high
- 2CVEs in the CISA KEV catalog
- 23public exploits
- Scannmap in Kali NetHunter, fixed profiles only.
- Read-only layerswhatweb, sslyze, ssh-audit, dig, p0f, UPnP. No logins, no exploits.
- CorrelationNVD, OSV.dev and CISA KEV. Exploit-DB as metadata only.
- AI write-upGemini explains each finding after pseudonymisation; a validator rejects made-up CVEs.
- ReportPDF: remediation plan, evidence and a confidence level for every result.


The report has a remediation plan (now, soon, while you are there), scan evidence and verification steps. Rysa itself points out that a human has to confirm the findings.
A Flutter (Dart) app, a Python (FastAPI) backend, over a thousand automated tests.
How Rysa works: technical details
- Scan. nmap in Kali NetHunter (a chroot on a rooted phone), commands built only from fixed profiles: no free-form flags, NSE scripts only from safe categories, and a target outside the allowed list needs a confirmation that I have the right to scan it.
- Read-only layers. whatweb (site technologies), sslyze (TLS), ssh-audit, dig (DNS resolver), p0f (passive OS fingerprinting), UPnP read from the router and the phone’s Wi-Fi security. No login attempts and no exploiting.
- Correlation with databases. Detected versions are matched against NVD, OSV.dev and the CISA KEV catalog (CVSS, vulnerabilities exploited in attacks). Exploit-DB is used only as metadata: I check whether an entry exists and never download its content.
- A description from an AI model. A Gemini model writes in Polish why a finding has its severity and how to check it without intrusion. Before that, IP addresses, MAC suffixes and host names are replaced with placeholders, and a separate validator rejects invented CVE numbers.
- Report. A PDF with a summary, a remediation plan, evidence and a list of gaps in the reconnaissance. Every result has a source and a confidence level, so a confirmed vulnerability is not mistaken for a match by version number.
nmap (Kali NetHunter) → whatweb, sslyze, ssh-audit, dig, p0f, UPnP → NVD, OSV.dev, CISA KEV, Exploit-DB (metadata) → a Gemini model with pseudonymisation and a CVE validator → PDF report.