Vulnix Agent

Chat with an AI pentester

Describe a target you own and an autonomous agent plans the attack, tests it in an isolated sandbox, and shows the whole thing live — the plan, the HTTP traffic, and every finding proven with a working exploit.

Start a scoped trial
Test the login and checkout on app.acme.dev — I own it.
Threat modelapp.acme.dev

A Next.js storefront with OAuth2 login and a coupon-gated checkout. Trust boundary at the payment callback; a self-registered customer is the attacker to model.

Scope decisions2 of 3 answered

How should I obtain test accounts?

Register fresh accounts myself with the disposable mailboxRegister two, for cross-account IDOR proof
Created plan3 phases · 9 tasks
  1. Done Recon & mappingEnumerate endpoints and auth flows
  2. Active Auth & business logicForge a coupon · test checkout IDOR
  3. Next Validation & report
Finding — confirmedHigh

Coupon signature is forgeable at the checkout callback

The HMAC secret is predictable; a crafted coupon=FREE100 is accepted and the order total drops to $0. Proven with the exact request and response below.

A pentester that plans, tests and proves — while you watch

01

You describe the target

Point the agent at a domain you have verified or a repository you have connected, and say what to test — the login, the checkout, the whole app. It only ever touches a target you own.

02

It agrees a plan first

The agent does its own recon, works out what the product is and where its trust boundaries sit, and lays out a plan you approve before anything is tested. No sandbox starts until you do.

03

It tests in an isolated sandbox

A fresh, network-isolated sandbox runs the whole engagement. You watch the plan move, the HTTP traffic stream, and each surface close, live — and the agent asks you when a decision is genuinely yours to make.

04

Every finding is proven

The agent reports a vulnerability only with a working exploit and the exact request and response behind it. No finding is inferred from a banner or a status code. Ship a fix and it re-runs the exact exploit to confirm.

Ask it the way you would a teammate

“Test the login and checkout on app.acme.dev”

A plan, then a live engagement across auth and business logic, with proof for anything it confirms.

“Is the coupon code on checkout forgeable?”

Focused testing of one surface — and an answer backed by the request that proves it, or a clear not-vulnerable.

“Review this repo for auth bugs before we ship”

A whitebox pass over your connected repository: the code paths, the entry points, and findings tied to the lines.

“Dig deeper into that IDOR and get me a stronger PoC”

The agent resumes the same workspace and goes further on a finding it already has — nothing is re-done from scratch.

How Chat works

The plan, the live workspace tabs, what the agent can and can't do, and how usage is counted.

Read the Chat guide

Blackbox and whitebox

Test a live app, a connected repository, or both — the agent runs the right methodology for each.

Blackbox vs whitebox

Questions

What is the Vulnix agent?
An autonomous AI pentester you talk to in plain language. You describe a target you own; the agent plans the engagement, tests it in an isolated sandbox, and reports each vulnerability with a working exploit. It is the same engine that runs Vulnix's scheduled pentests, in a conversation.
How is this different from a scheduled pentest?
Same engine, different entry point. A scheduled pentest runs the whole surface on its own; the chat agent is for when you want to steer — a question about one endpoint, a focused test, or to push further on something it already found. You watch and direct it as it works.
Can it attack anything I type?
No. The agent only tests targets you have proven you control — a verified domain or a connected repository — and that list is fixed for the whole conversation. Anything it reads while testing is treated as data to analyze, never as an instruction to widen the scope.
Does it make up findings?
No. A finding is reported only with a working exploit and the exact request and response that prove it. The agent treats everything the target returns — a 200, a 403, a 'not vulnerable' message — as a claim to corroborate, never as proof.
Does chatting use a pentest?
No. Chat is its own paid feature with a monthly usage allowance on each plan, separate from pentests. Planning is free; the agent's work counts against that allowance, which the sidebar shows as a percentage. Free plans don't include Chat.
Where does the agent run?
In a fresh, network-isolated sandbox created per engagement and torn down after. Credentials are injected per run and never baked into an image, and the request and response bodies you see are redacted of secrets before they leave the sandbox.