How the platform is designed

A sentence that starts with PLACEHOLDER is waiting for A to Z to confirm it. The architecture below is the design we are building to.

Our commitments

These are the rules the platform is designed around. They describe the intended service. They are not a statement that every part is switched on today.

• A client's documents, chats, memory, and tool connections live in one regional vault, chosen at sign-up.
• That content is isolated at the database level, with a dedicated encryption key for each client, and a tamper-proof activity log of every action.
• Agents read and prepare work. Nothing happens in a client's tools until a person approves it.
• Each company has its own account, with team invites and roles, so people do not share a password. User identity data stays inside our platform.
• Payments are handled by Stripe.

Data residency

The platform is designed as a central control plane plus separate regional data vaults.

The control plane holds accounts, login, billing, the agent catalogue, and the record of where each company's data lives. It does not hold the client's documents, chats, memory, or tool connections. Those live in one vault.

Launch regions are the United States and the European Union. The European Union vault is in Frankfurt.

A company outside those two launch regions, for example in the Middle East or in Africa, is hosted in the European Union vault under a data processing agreement.

Each region runs with a live standby copy, so that region keeps a second live copy for resilience.

Security controls

Inside a vault, the design separates each client and records what happens.

• Client data is isolated at the database level.
• Each client has a dedicated encryption key.
• Every action is written to a tamper-proof activity log.
• Identity is a company account. Admins invite teammates and give them roles. User identity data is kept within our platform.
• Planned connections are file upload, Odoo, Google Drive, Outlook, Xero, QuickBooks, and Zoho. Each one is designed to use OAuth and least privilege. They are planned. This page does not say each one is available today.
• Payments are handled by Stripe.

This website

The pages you are reading are the public marketing site. They are already served over HTTPS. The host adds these response headers on each page. This is separate from the vaults above.

• Content-Security-Policy. The browser may load scripts only from this site. Styles and fonts may also come from Google Fonts, because the pages use the Inter typeface. Drawings on the page use small style attributes, so styles allow inline rules. Scripts do not. The contact form uses a small inline image for the menu arrow, so images may also be a data URL from this site.
• Strict-Transport-Security, including subdomains, with a two year max age and the preload flag.
• X-Frame-Options set to DENY, and frame-ancestors set to none, so another site cannot frame these pages.
• X-Content-Type-Options set to nosniff.
• Referrer-Policy set to strict-origin-when-cross-origin.
• Permissions-Policy with camera, microphone and geolocation turned off.
• The website assistant and the contact form call this site only (/api/assistant and /api/contact). Anthropic is called by our server when an API key is set, not by the browser.
• Paths under /internal/ are left out of the sitemap, disallowed in robots.txt, and ask search engines not to index them. Keys for the assistant, for email, and for the enquiry store are environment variables on the host. They are not written into the pages.
• PLACEHOLDER: the preload flag is in the header. Sending the domain to the browser preload list is a separate step, and it is hard to undo. Confirm that every subdomain serves HTTPS before that submission.

Human approval

AI agents only read and prepare work. Nothing happens in a client's tools until a person approves it.

The agent can draft and prepare a summary for review. It does not send, file, change, or pay in a connected tool on its own. A person from the client account takes that step.

AI and your data

The platform is designed to be powered by Anthropic's Claude. The model is chosen for the size of the task. Regulated agents, covering tax, legal, HR, CFO, and due diligence, never run on the lightest model.

For finance agents, calculations are done by deterministic code. The AI explains the results. It does not replace the calculation.

Anthropic's commercial terms say that Anthropic may not train models on Customer Content from the Services. Those terms are at anthropic.com/legal/commercial-terms. Anthropic's privacy center says the commercial API default is that inputs and outputs are not used to train models, unless the customer opts in or sends feedback that allows it: privacy.anthropic.com.

Chat on this marketing site is a separate path. It is sent to the Anthropic API only when an API key is configured, and only from our server. This assistant does not send a chat rating to Anthropic.

PLACEHOLDER: confirm the Anthropic account for this site has not been switched on for model training.

The intelligence page publishes this line: Member of the Claude Partner Network.

Subprocessors

Two lists. The first is the platform design, for when the marketplace is in service. The second is this website, as it runs today. The longer account of the website is in the privacy notice.

Platform

Named here only where the design already names the provider.

• Anthropic, for Claude.
• Stripe, for payments.
• Odoo, Google Drive, Outlook, Xero, QuickBooks, and Zoho are planned connections to systems the client already uses. A person connects them with OAuth. They are not the hosts of the vaults.
• PLACEHOLDER: name the companies that host the control plane and the regional vaults. None are named on this page.

This website

These providers process personal information for a-zconsultancyltd.com when the relevant service is switched on.

• Vercel, for hosting and server logs.
• Resend, for enquiry email when it is configured. The default recipient is info@a-zconsultancyltd.com.
• Supabase, for the consultation_requests table when it is configured. If it is not configured, the email is the record.
• Anthropic, for assistant replies when an API key is configured. The browser does not call Anthropic. Our server does.
• Google Fonts, for the Inter typeface.
• PLACEHOLDER: Google Workspace, as the expected mailbox for info@a-zconsultancyltd.com, until that is confirmed or corrected.

Responsible disclosure

If you find a security problem, email info@a-zconsultancyltd.com. Please let us look at it before you describe it in public.

A machine-readable file is published at /.well-known/security.txt. It lists that email address, an expiry of 14 October 2027, the canonical URL, and English as the preferred language. The policy link in the file points back to this page.

Contact

Security questions: info@a-zconsultancyltd.com.

Privacy questions use the same address. The notice is on the privacy page.

To hear when the Agent Marketplace opens, use the contact page.