Security · Prolectio

How our audits read a website

Public pages only. Audits read what any anonymous visitor can see. We never log in, and we never audit behind a password.
Identified, not disguised. Our fetcher sends a User-Agent that names Prolectio and links back to this site.
Guarded fetching. Every address is resolved and checked before we connect: private, loopback, link-local and reserved networks are refused, redirects are followed one hop at a time and re-checked, and every request has a timeout and a size cap.
Read-only agents. In the deep audit, AI agents browse your site in a real browser to complete customer tasks, but the browser blocks every request that isn't a plain page load. Agents cannot submit forms, place orders or book appointments, and they cannot reach private networks.
Your site's text is treated as data. Page content passed to AI models is wrapped and labelled as untrusted, so instructions hidden in a page can't redirect the audit.

Client data and connections

Locked-down database. Every table has row-level security on and all public access revoked. Only our server-side systems can read or write data, and never directly from your browser.
Connections you can revoke. When you connect Google Calendar, Google Search Console or a Shopify store, we ask for the narrowest access the feature needs (calendar events and free/busy, read-only Search Console, read-only products, inventory and orders). The access tokens are stored encrypted in a secrets vault, never in ordinary tables, files or logs, and you can revoke them from your Google or Shopify account at any time.
We never hold your payment keys. The commerce kit sends buyers to Stripe payment links you create in your own Stripe account.
Google review text is kept for at most 30 days, as Google's terms require; after that we keep only dates, ratings and topic tags.
Card data stays with Stripe. We never see or store full card numbers.
Private links. Audit reports, concept previews and client pages use long unguessable links and are excluded from search engines.

Redesign concepts

Concept homepages are generated by AI from your site's own content. Before a concept is ever served, it is cleaned of scripts, event handlers, forms, embeds and external redirects.
Previews are then served under a strict browser security policy that allows no scripts and no form submissions, and runs the page in a sandbox.
A person reviews every concept before a client or prospect sees it.

Keeping AI costs and actions bounded

Every AI feature we run for clients has a hard spending cap in code, per audit, per concept or per client per month, so a bug or an abusive input can't run up costs. Our own site assistant is limited per visitor and alerts the founder when daily spend passes a threshold.
The site assistant can't take payments, issue refunds or change orders. It shows order status only after you enter a one-time code emailed to the address on the order, and payments happen only on our Stripe checkout page.
Nothing is sent to your customers, and nothing changes on your website, without a human approving it. Outreach drafts in Lead Generation are drafts for you to send; we never send outreach on a client's behalf.

How we build and ship

Every change goes through a pull request with automated type checks, linting and tests, plus a secret scan that blocks any committed key.
Secrets live only in our hosting and CI secret stores, never in code or chat.
Production changes, spending and anything client-facing require the founder's approval.

Reporting a problem

Found a vulnerability or a mistake in a report? Email contact@prolectio.com. We reply within one business day, and we'll correct or remove an audit or concept of your website on request. Current service health is on the status page.