Generation
Turn on Litestore's AI features with your own provider key, and review every draft they write before it goes live.
Litestore can use AI to draft text and suggestions in the admin. Use this page to see which features exist, turn them on, and learn how Litestore makes sure nothing goes live without your approval.
Litestore supports three AI providers: Google Gemini, Anthropic Claude and OpenAI. It connects to all three through one shared layer, the AI SDK. You pick one provider and add your own key, and a small set of admin features starts working. Each of them fills something in for you to review. None of them changes the storefront on its own.
Setup
Provider key
Add it to your environment. Set AI_PROVIDER if you're not using Google.
Connection test
Open Settings → AI and use the test button. The panel shows the active provider, the model, the fallback and this month's AI spend.
Assistant permissions
Leave the ceiling at read-only, or raise it when the assistant should prepare changes. Litestore never offers anything above the ceiling to the model.
Drafts
Use Fill with AI on the product page, the reply draft in Messages and the AI tag suggestions on the Tags page, and ask the assistant on the admin home for everything else.
Providers
Only one provider is active at a time. Each has a default model and its own key.
Google is the default. If you only set GOOGLE_GENERATIVE_AI_API_KEY, you don't need to configure anything else.
Google (default) gemini-2.0-flash GOOGLE_GENERATIVE_AI_API_KEY| Variable | What it does | Default |
|---|---|---|
AI_PROVIDER | Chooses the provider. Add the matching key with it. | |
AI_MODEL | Sets a different model from the same provider. | The provider's default model |
AI_FALLBACK_ENABLED | Turns the failover chain off when set to false. | — |
Failover
If the provider is rate-limited or returns a server error, Litestore retries the request with the next provider in this chain:
Google → Anthropic → OpenAI → GoogleLitestore only moves to a provider whose key you've also set. Failover doesn't add a second provider on its own.
If a key is rejected or revoked, Litestore never falls back. A broken key shows up as a failure instead of being hidden by a quiet switch to another provider. Otherwise, a dead key could go undetected for months while another provider bills the store for every request.
AI features
Fill with AI only writes into fields that are still empty, and specifications and tags are merged with what's there. Nothing is saved until you save the form. Every number it writes, such as a size or a weight, is checked against what you entered: a specification with a number that isn't in your inputs is held back for review instead of filled in, and copy with one is filled in but listed for you to check.
Product import
Once you've configured a provider, pasting a product URL into the import dialog pulls the product's content into a draft. Without a provider, the import dialog hides that option instead of letting it fail when you try it.
Approvals
The assistant can read the store's data directly, but it can't change anything on its own. When it proposes a change, Litestore stores the proposal, with the exact values, as a record and shows it as an approval card. The change only runs after you click Approve, and it runs with the values on the card.
you ask
↓
assistant reads store data directly
↓
change-making tool stores a proposal
↓
you click Approve within five minutes
↓
the change runs, onceMoney, refunds and stock have an extra safeguard. The code only allows proposals for them. A tool that tried to act on those directly would refuse to load. This isn't a setting that a later change could weaken by accident.
Settings → AI records every proposal and whether it was approved, dismissed or expired.
Storefront pages
The assistant can also write storefront pages. When you ask it for a new landing page or a change to an existing page, it first reads the existing pages and feeds, then proposes the page with its content, its product feed and whether it should be live. The page is only saved when you approve the proposal, and Litestore cleans up the HTML the same way it does for pages you write yourself. The assistant can also propose a search-result title and description for a page, product, collection or category. See Pages.
Limits
Two limits keep AI costs in check. Each person can make 20 AI requests a minute, and the store has a monthly AI budget, $50 by default. Once the month's spend reaches the budget, AI requests are refused until the next month, and Settings → AI shows the spend so far. Whoever runs the store changes both limits in the store's configuration.
Autonomy ceiling
Settings → AI has an autonomy ceiling for the whole store, set to read-only by default.
If a capability is above the ceiling, Litestore never shows it to the model. The model is never offered the capability and then blocked from using it. The capability isn't in its tool list, so nobody can talk it into using it.
Input screening
Before a prompt reaches a provider, Litestore checks it for injection patterns. If Litestore blocks a message, it drops it instead of passing it on.
AI features stay hidden until you add a provider key
Until a provider key is set, the admin home shows a Set up AI button instead of the assistant, and the admin hides Fill with AI, the reply draft in Messages, the tag suggestions and the URL product import. The AI endpoints return 501.
Scope
| Capability | Included | Required activation |
|---|---|---|
| Assistant questions with tables and charts | Yes, on the admin home | A provider key |
| Changes prepared by the assistant (refunds, discount codes, campaigns, prices, stock, customer replies) | Yes, as proposals you approve | A provider key, plus the ceiling at Read + Propose |
| Content changes prepared by the assistant (pages, feeds, product descriptions, search titles, translations, broadcasts) | Yes, as proposals you approve | A provider key, plus the ceiling at Read + Propose + Execute |
| Fill with AI on a product | Yes, as a prefill you save yourself | A provider key |
| Reply drafts in Messages | Yes | A provider key |
| Tag suggestions | Yes, on the Tags page | A provider key |
| Product import from a URL | Yes, in the import dialog | A provider key |
| Image alt text generation | No, not included | — |
| Bulk generation across a collection | No. Each generator works on one record | — |
Related
Conversations
Answer customer questions in the Messages inbox, attach products, orders and coupons to replies, and leave internal notes for your team.
Checkout
Litestore prices the cart, holds stock while the shopper pays and creates the order once payment is confirmed, and a few checks help when checkout fails.