← All projects
Security

StrixOC

Autonomous AI penetration testing

Overview

StrixOC runs your code dynamically, finds vulnerabilities, and validates them through real proofs-of-concept. Multi-agent orchestration lets a team of AI pentesters collaborate and scale. Used internally as a reference implementation for SecOpent's adapter path and as a fast-feedback channel for new case-DSL primitives before they land in production.

GitHub Actions / CI integrations and a developer-first CLI keep the loop short. Open-source LLM providers supported; no vendor lock-in.

Product impact

Pain → Solution → Value

Penetration testing is sold as a once-a-year snapshot — expensive, slow to schedule, and stale the moment the application ships. StrixOC replaces the annual engagement with an autonomous agent team that runs per release, exploits what it finds, and outputs Pydantic-validated, signed findings.

Pain
Annual pentest = snapshot in time
App ships weekly; coverage is yearly.
Solution
AI agent runs per release
Nightly in CI, on every deploy.
24/7
Continuous vs 1× / year
Pain
Static-scanner noise
Thousands of unvalidated alerts.
Solution
Dynamic real exploitation
Agent proves the vuln before reporting.
↓" data-i18n-zh="">
False-positive cleanup
Pain
Expensive manual labor
Senior hours × weeks per run.
Solution
Token-cost agent
Minutes + tokens, not consulting.
↓" data-i18n-zh="">
$ per finding
Pain
Findings not actionable
Reports with no fixes attached.
Solution
Pydantic-validated + signed
Structurally sound, signed + fix proposed.
↑100%" data-i18n-zh="100%">100%
Actionable output

What changes for the team

A wider view of the same loop. Each card zooms into the corresponding column above.

Pain

Annual pentest is stale on arrival

  • Expensive, slow to schedule, dependent on senior-tester time on your attack surface.
  • Between engagements the app keeps shipping — new endpoints, deps, auth paths — none tested.
  • Static scanners fill the gap with noise: thousands of unvalidated alerts.
12 mo
Gap between pentests
Solution

Autonomous red team on tap

  • Agent teams (openai-agents SDK + LiteLLM) collaborate like a real red team — recon, exploit, report.
  • Operator-defined scope; every step validated against Pydantic — agents cannot drift.
  • Agents propose and generate fixes — not just findings.
100%
Findings structurally sound
Value

Continuous coverage, lower cost curve

  • Pentesting becomes continuous: a run costs LLM tokens + minutes, not a consulting engagement.
  • Time-to-discovery drops from months to the length of a single run.
  • Apache-2.0, self-hostable, no single-vendor LLM dependency — cost curve under your control.
1× / release
vs 1× / year engagement

Product architecture

01
Scope

Operator declares the boundary from the CLI.

02
Recon

Agent maps routes, params, auth flows, dependencies.

03
Exploit

Real exploitation attempts, inside safety rails.

04
Report

Pydantic-validated findings, ready for CI or tickets.

Every run moves through the same five participants.

Technical architecture

AI penetration testing · usestrix/strix fork. Click any node or relationship for details.

Capabilities

Real exploit validation
Real PoCs through dynamic execution, not the false positives of static analyzers.
Multi-agent teams
Collaborative pentesters that scale the way real red teams do.
Developer-first CLI
Actionable findings with remediation guidance, not 50-page PDFs.
Auto-fix and reporting
Generates patches and compliance-ready pentest reports.

By the numbers

  • Real PoC validation
  • Multi-agent orchestration
  • Developer-first CLI
  • Auto-fix + reporting
  • GitHub Actions integration
  • Open-source LLM providers

Try it, fork it, or hire me to extend it.

Source on GitHub · {p['license']} · questions to 286043314+echocc00@users.noreply.github.com