Run a Security Testing Agency with Agents
Agents test each client's live product as a customer would, push on its edges and write up every gap with steps to reproduce. You approve every session.
From $20/mo per company, includes $20 of credits. Pricing
How to Start an AI Security Testing Agency
- Team
- 5 agents
- Per client
- 2 testers
- Browser reaches
- The client's own site
- Sessions start
- Only after you approve
An AI security testing agency uses each client's live product the way a real customer would, then pushes on its edges. Every gap comes back with steps to reproduce. You approve every session.
What is an AI security testing agency?
It is a ready-made naïve company. One deploy hires an agency team and installs the default apps: a CRM for your clients, a calendar for deliverables and meetings, and the SEO and Ads apps for your own marketing. Each client gets a testing team of their own, hired by your CEO agent with your approval.
The agency team splits the work three ways:
- Sales. Finds and follows up with products that need testing, and drafts every email for you to approve.
- Client manager. Keeps each client's record, scope and deadlines true in the CRM and the calendar.
- Marketing. Writes your own site and posts, so prospects can see what you test and how.
Each client's team adds two testers:
- Customer tester. Walks the product's main journeys as a real customer would: sign up, buy, cancel, reset a password. It writes down every place the product gets in the way.
- Edge-case tester. Tries what customers do by accident: empty fields, long names, the back button, two tabs, a slow connection. It writes down every place the product breaks.
Three agencies it could run:
- Launch checks. A test pass on a startup's product the week before it opens to the public.
- Monthly regression. The same journeys, every month, so a client knows what changed.
- Checkout reviews. A focused pass on the steps where a shop takes money.
Set it up in four steps
1Deploy the template
Tap Deploy on this page. You make your naïve account on the way. The agency team and the default apps are installed together.
2Add your first client
Add the client to the CRM with their website domain. Then open the CEO chat and ask it to onboard them. The CEO reads which testers the template names for that client and asks you to approve each hire.
What should the first test pass cover?
The testers read this before every session, so they spend their time on the journeys that matter to this client.
- Sign up and log in
- Checkout and payment
- Account settings and cancel
- Everything a new customer sees in the first week
3Testers work the product
Once you approve a session, the customer tester walks the journeys in the scope and the edge-case tester pushes on them. Each gap is saved to that client's files with the page, what was done, what was expected and what happened instead.
4Approve before a report leaves
The client manager drafts the gap report from the saved findings. It waits for you with the findings attached. Read it, then allow or hold. Nothing is sent to the client before you allow it.
Client manager wants to send the checkout gap report
Six gaps in checkout, two that stop a payment.
Each one has the page, the steps to reproduce and what should have happened.
- Client
- Larkspur Outdoor
- Scope
- Checkout and payment
- Sends
- After you allow
What should an AI security testing agency sell?
One clear kind of test for one kind of product. Ten that work:
- Pre-launch passes. The main journeys, the week before a launch.
- Checkout reviews. Every step where money changes hands.
- Sign-up and login checks. Password resets, email links, locked accounts.
- Form testing. Empty, long, odd and pasted inputs on every field.
- Monthly regression. The same journeys, compared month to month.
- Mobile browser passes. The product on a small screen.
- Cancellation flows. Whether a customer can leave as easily as they joined.
- Permission checks. What a logged-out visitor can still see.
- Error page reviews. What a customer sees when something fails.
- Release checks. A short pass on the pages a new release touched.
Can an AI security testing agency make money?
It can, if you sell a clear scope and report only what you can reproduce.
| Model | How it earns | The catch |
|---|---|---|
| Fixed pass | A set fee for one product and one scope. | You must agree the scope before you start. |
| Monthly retainer | A set fee for a regression pass every month. | Quiet months still have to show value. |
| Release checks | A price per release tested. | Clients ship on their own schedule, not yours. |
naïve does not find clients for you. The team tests the products you bring, and Sales helps you reach more. A gap report is evidence of what happened in a browser. It is not a certificate that a product is secure, and it should never be sold as one.
What does your first week look like?
- Day 1. Deploy, add one client with their domain, and approve their testers.
- Day 2. Approve the first session. Read the findings as they are saved.
- Day 7. Send the first gap report. Book the next pass in the calendar.
Want a different agency? Try the AI Automation Agency or the Market Research Agency. See how human approval works for the mechanism underneath.
The week it runs
clients-security/agency · v0.1.0 · published Nov 5, 2026
2 timers are provisioned with this template — 6 unattended runs a week. Each fire opens a session with the message excerpted below and spends against its own cap, which the studio prints before you deploy.
- 08:30every weekday at 08:30Sales30 8 * * 1-5
Morning pass. This chat works for no client and keeps it that way: it reads the CRM's directory and sees who wrote, not any client's record or mail, so open no…
- 09:00every Monday at 09:00Client manager0 9 * * 1
Monday. This chat works for no client and keeps it that way, so open no contact here. Read this week's Calendar and your won contacts with crm.read. For each…
The Agent Team
read from the clients-security release
The 5 agents naive up provisions into your organization — the role each holds, the tools it is allowed, the skills it was taught, the timers it owns and the first thing it is told. An agent marked required cannot be left out of an install; the rest are yours to decline.
Sales
New businessrequiredFinds leads, writes to them, follows up and books the first call. Every email waits for your yes.
toolsask_operatorcrm.readcrm.writeemail.inboxes… 8 more tools
skills
1 timer·08:30
vetta/auto
Client manager
AccountsrequiredRuns each client: the kickoff, the calls, the weekly update and the renewal.
toolsask_operatorcrm.readcrm.writeemail.inboxes… 6 more tools
skillsnaive/qa-gap-report
1 timer·09:00
vetta/auto
Customer tester
Journey testingrequiredUses each client's product the way a new customer would, and reports where it gets hard.
toolsask_operatorbrowsercrm.readcrm.write… 12 more tools
skillsnaive/customer-journey-testingnaive/qa-gap-report
No timer — works when you ask it to.
vetta/auto
Edge-case tester
Edge casesrequiredPushes on the journeys the customer tester walked: odd inputs, odd orders and broken states.
toolsask_operatorbrowsercrm.readcrm.write… 12 more tools
skillsnaive/edge-case-testingnaive/qa-gap-report
No timer — works when you ask it to.
vetta/auto
Marketing
ContentrequiredWrites the posts and case studies that bring in leads, each one saved for you to approve.
toolsask_operatorcrm.readcrm.writeedit… 14 more tools
skills
No timer — works when you ask it to.
vetta/auto
You will be asked
read from the clients-security release
Answered in the studio before anything is provisioned, and read by every agent before it acts. Configuration only — a secret is a vault credential or an app secret, never an answer. You can change the answers later.
- 01
What does your agency do, and for whom?
Your team writes to clients and leads from this.
free text
- 02
What do you sell, and for about how much?
Journey and edge-case testing, $2,500 a month
- 03
What kinds of products do your clients ship, and which journeys matter most?
Your testers plan each client's journeys from this.
Web apps with a self-serve signup; signup, onboarding and the first paid action.
Skills
read from the clients-security release
The procedures the agent team loads before it works — what to read first, what to file where, what never to do. naive/ names the platform's own catalogue, which the studio's Skills screen carries; a bare slug is a skill the template ships itself.
What it ships
v0.1.0
A template is a repo: the apps below, their server, and a naive.config.ts declaring the agent team. Clone it, bind it to your organization, bring it up.
Questions
Can I deploy the security testing agency today?
Yes. Tap Deploy on this page. The team, the CRM, the calendar and the other default apps are set up in one step.
Is this a penetration test?
No. The testers use the product through a browser, as a customer would, and push on its edges: odd inputs, broken flows, missing checks. They do not attack servers or try to break in. A client who needs a formal penetration test still needs a firm that does that work.
Can a tester reach a site that is not the client's?
No. When the CEO hires a client's team, each tester's browser is locked to that client's own public domain. A client with no domain, a private address or a wildcard gets no team until that is fixed.
Can one client's findings end up in another client's report?
No. A chat works for one client from the moment it opens that client's record. It sees only that client's files, saves files tagged to that client, and cannot email another client's contacts.
Will the team test without me?
No. Every browser session waits for your approval first. Nothing runs against a client's product until you allow it.
Start this company
5 agents, 0 apps, a budget. Three ways in.
01 — Deploy from template
No terminal. Choose a template in the browser, deploy it, and watch it work.
naïve / studio
- Security Testing Agencydeploy
- Naive Short Form v1deploy
- Naive Long Form v1preview
- Naive Clipping v1deploy
- AI Music Channeldeploy