The platform

An AI pentester that proves every finding

Verify a target, let an autonomous agent attack it in an isolated sandbox, and keep only what it can exploit. Then fix, re-test and review every pull request with the same engine.

Start a scoped trialSee pricing

How it works

From a verified target to a proven fix

  1. 01

    Verify what's yours

    Add a domain and prove you control it with a DNS record, a file or a meta tag, or connect a GitHub repository. Nothing unverified is ever tested.

  2. 02

    The agent attacks it

    An autonomous agent maps the target and tests it in a fresh, network-isolated sandbox, from the outside, from the source, or both.

  3. 03

    Only proof survives

    A finding is kept only with a working exploit and the exact request and response behind it, ranked by severity with its CWE.

  4. 04

    Fix it, then prove the fix

    Track it to Jira, open a fix pull request for whitebox findings, and replay the original exploit against your patch before you close it.

The product

Every way you test, on one platform

Each part runs on the same engine and lands in the same findings list.

Blackbox & whitebox pentests

Test a live app or API like an attacker, read the source first, or both in one run.

How the two compare

Vulnix Agent

Chat with the pentester: agree a plan, watch it work, and steer it as it goes.

Meet the agent

PR reviews

Every qualifying pull request reviewed on GitHub before it merges, inline.

See PR reviews

Findings & reports

One ranked list of every proven finding, and a PDF report for each test.

Explore findings

Fix validation

Replay the exact original exploit against your patch. A real re-test, not a checkbox.

How Validate Fix works

Vulnix MCP

Start runs and read findings from Claude, ChatGPT or Cursor, signed in with OAuth.

Connect your AI tools

Proof, not guesses

Every finding comes with the exploit that proved it

The agent treats whatever the target returns as a claim to check, never as proof. A vulnerability reaches your list only once it has been exploited, so there is nothing to triage away.

  • The exact request and response behind each blackbox finding
  • The file and line for each whitebox finding
  • Severity, CVSS and CWE on every one

Trust

Built to be pointed at production

A sandbox per run

Every engagement gets its own network-isolated sandbox, created for the run and torn down after it.

Verified targets only

Runs start only against domains you've proven you control, connected repositories, or public IP ranges.

Credentials stay sealed

Test-user credentials and headers are encrypted at rest and decrypted only inside that run's own sandbox.

Connect

Fits the tools you already use

GitHub

Whitebox pentests, automatic PR reviews and fix pull requests from one GitHub App.

Jira

Create an issue from a finding, with its severity and evidence, in the project you choose.

Slack

Post finished pentests and new findings to the channels your team already watches.

Webhooks & API

Signed webhooks for finished pentests, PR verdicts and finding changes, and a REST API with scoped tokens, on Startup and above.

MCP for AI apps

A hosted MCP server: your AI coding agent can start runs and read findings, read-only by default.

Cloudflare & Vercel DNS

Connect your DNS provider and Vulnix adds the verification record for you.

Questions

What does Vulnix actually do?
It runs AI penetration tests. You connect a target, a domain you have verified or a GitHub repository, and an autonomous agent tests it inside its own isolated sandbox. It keeps only the findings it can prove with a working exploit, and can re-run that exploit later to check a fix. The same engine also reviews your pull requests.
Is this a scanner?
No. A scanner matches signatures and reports what might be wrong. The Vulnix agent plans an engagement, chains requests the way an attacker would, and only reports a vulnerability once it has exploited it, with the request and response that prove it.
What can it test?
Web apps and APIs on domains you have verified, public IP ranges (port and service testing of internet-facing infrastructure), and GitHub repositories. A whitebox scope can also carry a verified domain, so one run covers the live app and its source.
Do I need to sign an authorization form?
No separate form. Proving you control the domain, with a DNS TXT record, an uploaded file or a meta tag, is what lets a run against it start.
What does the free plan include?
One full pentest, blackbox or whitebox, and 5 PR reviews a month, with no card and no expiry. High and critical findings show as a preview until you choose a plan.