From FTP edits to an AI-managed pipeline in about a week
Two online stores, 42,000 products synced from 1C, and changes made directly on the live server. One week later, the owner was shipping changes by describing them to an AI agent.
AI-managed websites & shops
Change your website by chatting with AI, safely
You describe the change. An AI agent makes it. Automated checks, a staging preview and a self-checking deploy make sure it cannot break your shop.
We move your existing website or online shop onto a governed Git pipeline so that you, the owner, can ask an AI agent (Claude) for changes in plain language, preview them on a staging copy, and approve them, without hiring a developer for everyday work. Setup starts at €2,500 and usually takes about one week; an optional care plan starts at €300 per month.
Who it is for
- Owners and managers of online shops (OpenCart, ocStore, WooCommerce, Magento, Laravel and custom PHP)
- Companies whose website changes wait weeks for a developer
- Teams already trying ChatGPT or Claude for site edits and worried about breaking production
Why now
- Every small change costs a developer invoice and a week of waiting
- Edits are made straight on the live server over FTP, with no history
- You have tried an AI tool on the site and are not sure what it changed
- Your previous developer or agency is gone and nobody knows how the site is deployed
What you get
- Backups first: encrypted, off-site, database every 6 hours, with a tested restore
- Server hardening: origin only behind the CDN/WAF, PHP only for allowed entry points, SSH keys only, FTP removed
- Code in Git in your own private GitHub organisation, after a scan for leaked secrets
- Staging copy on sanitised data: payments, email, analytics and social login switched off
- Automated checks on every change: syntax for your server's PHP version, secrets, forbidden files, and a flag for risky changes
- Production gate on the server: only reviewed pull requests with green checks deploy; backup before, health check after, automatic rollback
- Nightly drift check that opens a ticket if anyone edits production outside the pipeline
- Rules for the AI agent (a
CLAUDE.mdwritten for your site) and a 10-minute setup for the owner
How it works
Look and report
We inspect the server and code read-only and send you a plain-language report of what is risky today.
Make it safe
Backups, hardening and code in Git before anything else changes.
Build the pipeline
Staging, checks and the production gate, tested with real changes.
Hand over
You make your first change with us on a call. From then on you request, preview and approve.
What changes for you
Before, a change meant writing to a developer, waiting, paying, and hoping nothing else broke. After, it looks like this:
- You open Claude and write: “Show the warranty period under the price on product pages.”
- The agent records the request, makes the change on a separate branch, and opens a pull request.
- Automated checks run. If the change touches something risky, it is flagged for an engineer.
- The change appears on staging, a private copy of your shop, and you get the link.
- You look, and say “approve”.
- The change is deployed. The server takes a backup first and checks the site afterwards. If something is wrong, it rolls back by itself.
Every request, decision and result stays in the history, so in a year you can still see who changed what and why.
Why this is safe enough for a real business
This is the same change control we run for payments infrastructure, scaled down. The important part is that the rules live on your server, not in the AI. Even a perfectly obedient agent, a careless developer or a compromised laptop cannot push code that did not pass review, checks and the gate.
The AI agent has no access to your server, database or passwords. It can only propose changes.
Related work
A safe delivery pipeline for a 9-year-old marketplace, live in one day
A Laravel marketplace with about 3,900 commits since 2017, one dedicated server under memory pressure, and changes going straight to production. The gated pipeline went live in a single day.
Questions clients ask
What kind of changes can I make myself?
Text and translations, layout and styling, how products are displayed, banners, filters, small features, SEO fixes and markup. Anything touching payments, the ERP/1C exchange, database structure, server configuration or deleted data is automatically flagged and needs an engineer's approval.
Do I need to know Git or programming?
No. You install the Claude app and the GitHub app, which takes about 10 minutes and which we do with you. Then you write requests in plain language. The agent handles branches and pull requests and sends you a staging link to check.
What stops the AI from breaking the live site?
The AI never touches the live server. It can only open pull requests. Production accepts a change only if it came through a reviewed pull request with green checks, and the gate script on the server makes a backup, deploys, checks that the site responds, and rolls back on its own if it does not.
Is my code and data mine?
Yes. The code lives in your own GitHub organisation, the servers stay yours, and staging uses a sanitised copy. We keep no copies after the engagement.
Which AI do you use?
We set it up for Claude by Anthropic, because its coding agent works well with rule files and pull requests. The pipeline itself does not depend on any AI vendor; a human developer goes through exactly the same gates.
What if my site is not on OpenCart?
The approach works for any site whose code can live in Git: PHP, Laravel, WordPress/WooCommerce, Magento, Node.js, static sites. The checks are adapted to your stack.
See it on your own shop
Send us your site address. We will tell you what the setup would involve and what is risky today, for free.