The answer lives in another system
Stock in the ERP, traffic in analytics, orders in the store. Cross-referencing that is manual work nobody does every week.
Castle Copilot is an AI assistant wired to your e-commerce, your ERP and the tools you already use. You type or send a voice note; it queries the data, shows the answer and executes whatever you authorize.
Included in the monthly plan for clients we operate with
White-label dashboard with an assistant that queries store, ERP and campaigns — and runs tasks in natural language, with your approval.

Under construction
Castle Copilot is in development. The capabilities below describe the product we are building — in the conversation we tell you exactly what is available today for your operation.
Your operation already produces everything you need to know. The problem is the answer is spread across four systems and depends on someone finding time to build the spreadsheet.
Stock in the ERP, traffic in analytics, orders in the store. Cross-referencing that is manual work nobody does every week.
Product descriptions, price reviews, checking stuck orders — same work, always late.
By the time the number shows up in the month-end close, the decision it called for has passed.
Ask, query, propose and execute — with your approval on every step that changes data.
In the browser or on WhatsApp, typing or sending a voice note. No menus, no report to configure.
Store, ERP, shipping, payments and analytics — the same integrations Castle built on your project.
Lookups come back ready. Changes come back as a proposal, for you to approve before anything moves.
The action is logged, with what changed and when — auditable afterwards.
Example · weekly sales


It arrives already connected — because we are the ones who integrated your store.
Open architecture
Copilot runs on MCP servers — connected or custom-built for each client’s systems.
Model Context Protocol (MCP) is the standard that links AI assistants to real data and actions. Instead of pasting spreadsheets into chat, your systems become tools any compatible model can query and run.
Chat in the browser or WhatsApp: ask, see data and approve actions — nothing to install.
Connect ChatGPT, Gemini, Claude or any MCP-capable client — same servers, same operation.
Servers wired to your systems — store, ERP, shipping, campaigns — built and maintained by Castle.
Switch model or interface without redoing integrations — value lives in MCP connectors, not the assistant of the month.
Claude, Gemini or ChatGPT — any MCP-compatible LLM reaches the same servers in your operation.
Switch models without redoing integrations — same question, same systems connected via MCP.
Organized by the risk of each action — because reading a number and changing a price are not the same thing.
Read-only. It looks things up and answers, changing nothing.

It generates the content; nothing publishes until you say so.

It changes data in your system — always with your approval.

It monitors on its own and alerts you when something drifts.

No action that changes data happens on its own. It proposes, you confirm.
Every lookup and every execution is recorded, with author, timestamp and what changed.
Who can ask, who can approve and what each person sees is defined by you.
Your system credentials stay in your environment and do not feed a third-party model.
Ask and draft, with a monthly usage allowance, for every client on continuous follow-through with Castle.
Act and watch — actions that write to your systems, plus continuous monitoring — come as a separate module, sized with you.
No fixed price list: the number comes out of the conversation, along with the scope of your follow-through.
Tell us which systems you use today. We reply with what can be connected now and what comes later.
We reply during business hours