1
Activate
Switch the "Activate" toggle. As long as it's off, no API calls are possible — even if a key is configured.
Documentation · 06
AI text generation with ChatGPT & Claude.
Generate product texts automatically — optionally with ChatGPT (OpenAI) or Claude (Anthropic), both usable in parallel with provider selection per generation.
Provider
Both configurable at once.
Own API keys, own model choice, own tone defaults. You pay the provider directly — Stone & Water earns nothing on token usage.
C
Claude
Anthropic
Strengths for longer, nuanced texts and brand voice. Fully configurable independently of ChatGPT.
Models: Haiku 4.5 · Sonnet 4.6 · Opus 4.6 / 4.7
G
ChatGPT
OpenAI
Proven for fast, cost-effective generation of large volumes. Own key/model/tone configuration. Connection test directly in the admin.
Models: GPT-4o Mini · GPT-4o · GPT-4.1 Mini · GPT-4.1
Setup
In six steps per provider.
Each provider has its own section in the config matrix under "Translations". Both sections work analogously.
2
API key
From platform.openai.com (sk-...) or console.anthropic.com (sk-ant-...). Stored encrypted in the database.
3
Select model
Haiku/Mini cheap & fast, Sonnet/4o as default, Opus/4.1 for highest quality at higher cost.
4
Tone default
Professional, casual, premium, technical or emotional. Overridable per call.
5
Enable bulk
"Enable bulk" toggle — otherwise the provider can only be used in the detail modal, not in bulk-edit.
6
Test connection
Test button sends a 50-token ping. Success/error is displayed. Reasonable sanity check before productive use.
Generator modal
In the product detail above the description field.
Button "AI texts" opens the modal with all relevant options — fields, languages, tone, context.
Fields
What is generated:
Description (HTML), meta title (max 60 chars), meta description (max 155 chars), keywords (comma-separated) — individually or together.
Languages
Pick from active
Any number of languages. Optionally: individual texts per language, or generate once and clone.
Tone
5 style templates
Default comes from the config matrix, overridable per call. Visible in the preview header.
Context chips
You control what the AI sees
Product name, manufacturer, properties, category, price, selected custom fields — selectable per chip. Plus your own SEO keywords.
Provider switch
When both active
Appears only when both ChatGPT and Claude are configured. Selectable per call. Defaults to the last used.
Preview
Review before applying
Generated texts first appear in preview mode. Review them, edit manually — then "Apply". "Overwrite existing" is optional.
AI in bulk
Texts for many products at once.
When a provider has "Bulk enabled", it can also be used in bulk-edit. Every product gets its own text based on its data.
Cost warning
AI bulk can quickly get expensive.
1000 products × 1500 tokens per text × Sonnet 4.6 price = a few euros. Plan bulk runs carefully — and use the automatic snapshot in case you want to roll back. Concurrency control in the config matrix prevents rate-limit issues at scale.
Privacy
You decide what goes out.
On every AI call, the selected product data is sent to the respective provider (OpenAI / Anthropic). If you don't want that, disable the provider — no external API calls happen then.
Proxy
Server-side, not browser-side
The PIM proxies calls through your Shopware server. OpenAI / Anthropic see your server IP, not end-customer IPs.
Keys
Encrypted in DB
API keys are stored with Shopware's own encryption logic. They are visible in the UI only as "••••".
Context control
Controlled per chip
You control which product data flows into the prompt. Sensitive CFs (e.g. purchase prices) shouldn't be actively selected.
Billing
Directly with provider
You pay OpenAI / Anthropic directly. Stone & Water earns nothing on token usage. No reseller model.
Questions?
We help with onboarding.
Personal setup, data migration from legacy systems, training for your team — on request, we get the PIM running productively in your shop.