FAQ Kit AI Generator

Writing FAQs is slow when a catalog is large: every product of a kind needs the same sort of questions, and each answer has to match that product. Drafting them in a chatbot means copying text back and forth and checking every line.

FAQ Kit AI Generator puts that drafting inside the admin of Adobe Commerce and Magento Open Source. From a product, a category, a FAQ widget or the FAQ Entries grid, it generates candidate questions and answers from the item's own catalog content or from text you supply, lists them for review, and saves only the ones you keep as FAQ Kit entries. It is free, and it is the AI part of FAQ Kit – FAQ Kit itself curates, renders and serves the FAQs.

• You run FAQ Kit and want first drafts for many products or categories instead of writing each FAQ from a blank page.
• Your product descriptions are good, and the FAQs should be answered from them rather than from a model's general knowledge.
• You have a policy page, a supplier sheet or a support macro as text, and want store-wide or widget FAQs drafted from it.
• You already use LLM Provider, or want AI features that run with your own provider key, a model you choose and a daily budget you set.

• You do not use FAQ Kit. This package only adds generation to FAQ Kit; without it there is nothing to generate into, and Composer will install FAQ Kit as a requirement.
• You do not want a model provider account. Generation needs a chat model configured in LLM Provider, and the models this package lists are Anthropic and OpenAI models used with your own API key. Nothing is resold here.
• Your source content must not leave the server for a commercial provider. The source text of each run is sent to the provider you chose. LLM Provider can also reach a local Ollama server, but a local model is not on this package's tested list: it has to support structured JSON output and be admitted with Allow Untested Models. Otherwise write the FAQs by hand – FAQ Kit does not need this package.
• You need FAQs written and published without a person looking at them. This package never saves a candidate on its own.

• Four starting points. Generate with AI on the FAQ Entries grid, and Generate FAQs with AI on the product, category and FAQ widget edit forms. The forms open the review page already scoped to the item; the grid lets you pick global, product, category or widget and the store view.
• Grounded candidates. For a product: its name, SKU, short description and description. For a category: its name and description. On top of that, or as the only source for global and widget FAQs, text pasted into the page or a .txt or .md file uploaded there (up to 2 MB, read up to 40,000 characters, not stored).
• Review, then apply. Every candidate is a ticked row. Generate more adds further candidates without repeating the listed ones; questions that already exist for the same target are filtered out. Apply Selected saves only the ticked rows, through FAQ Kit's answer sanitising, assigned to the item you started from, with the time of generation recorded on the entry.
• Your settings. The prompt, pairs per run, language (or the store view's locale), question and answer length caps and an answer style are FAQ Kit settings. The model and a daily spend cap for FAQ generation are chosen under LLM Provider → Consumer Modules → FAQ Kit, from chat models that support structured JSON output.
• Built to fail soft. Output is constrained to a JSON schema of question and answer pairs, a retryable provider error is retried once, and any failure is reported on the review page while the rest of the admin keeps working. Every call appears in LLM Provider's usage log with tokens and estimated cost.

• It does not work alone. No generation without FAQ Kit and LLM Provider, both required and both installed by Composer.
• It does not include model access. No provider account, key or quota comes with it; you pay your provider directly.
• It never publishes on its own. Candidates are saved only when you tick them and press Apply Selected. There is no scheduled, background or bulk generation across many products at once – one product, category, widget or store-wide run at a time.
• It reads plain text only. Sources are catalog content, pasted text and .txt or .md files – no PDF, Word or spreadsheet import, no web page fetch, and no product attributes beyond the four named above.
• It does not translate or rewrite existing FAQs. It drafts new ones in the configured language; editing stays in FAQ Kit.
• Costs are estimates. LLM Provider computes them from its price table, not from your provider's invoice, and the daily cap is a soft ceiling: the call that crosses it completes, the next one is refused.