Skip to main content
The @frontic/nuxt module wires your Nuxt application to the Frontic backend. It ships composables for data fetching, automatic proxy configuration, and locale management.

Installation

This installs the module and adds it to your nuxt.config.ts automatically.
nuxt.config.ts
The module works out of the box with sensible defaults: CORS proxy enabled, all composables registered, and TypeScript paths configured.

Configuration

All options are optional. Customize behavior as needed:
nuxt.config.ts

Options

API

boolean | string
default:"true"
Enable the built-in CORS proxy at /api/frontic, or set a custom path.
number
default:"10000"
Milliseconds before the proxy gives up on the Fetch API. Set 0 to disable the timeout.Without a bound, a stalled upstream holds the request until Node’s default of 300 seconds — long after the page is lost, and long enough to exhaust your server’s connection pool. A timeout returns 504 Gateway Timeout and fires frontic:fetch:error.
string
The secret for a project whose Fetch API is protected. Server-only — kept in private runtime config, never exposed to the browser.

Routing

boolean
default:"true"
Automatically redirect on 301 responses in useFronticPage.
boolean
default:"true"
Throw a 404 error when page is not found in useFronticPage.
boolean | { domains?: string[] }
default:"false"
Serve the project’s sitemap from your app at /sitemap.xml, with domains as the served domains. See Sitemap.

Context

boolean
default:"false"
Disable automatic context fetching and cookie management in useFronticContext.
Cookie name for storing the context token.
Cookie max age in seconds. Default is 1 year.
string | boolean
The value every composable sends as contextDomain, which Frontic matches against your domain rows to resolve scope, region, and locale. It also forms the host portion of the slug useFronticPage looks up.When not set, useFronticPage derives the domain from the request URL host (useRequestURL().host). This works in production where the host is the Frontic domain, but not when the host doesn’t match — for example in local development (localhost:3000), sandbox / preview environments, or staging where only the production domain is configured in Frontic.You can configure a domain alias in your admin app as an alternative. If your dev environment uses a host that isn’t configured as an alias, set contextDomain to override it:
Per-locale domains
A multi-locale storefront needs a different contextDomain per locale, because each of your domain rows binds one scope / region / locale triple. Setting contextDomain: true tells the module to read the domain off the locale that @nuxtjs/i18n currently has active, so it follows locale switches without any per-call wiring. contextDomain is not an i18n feature — it’s a custom property you add to each locale object, which the module then reads through i18n’s localeProperties:
Each value must match a domain row in Settings → Domains. With the rows above, a visitor on the de locale resolves to the German region and locale, and useFronticPage looks up demo-shop.com/de/produkte/schuhe for the path /produkte/schuhe.
contextDomain: true without @nuxtjs/i18n installed, or a locale object with no contextDomain property, logs a warning at build or request time and sends no domain — Frontic then falls back to the project’s default scope, region, and locale rather than erroring. Watch for that warning if a locale silently serves the wrong market.
Resolution order
Every composable resolves the domain the same way, taking the first that applies:
  1. The contextDomain passed in the composable’s own options — per call, and reactive.
  2. The module’s contextDomain string.
  3. The active i18n locale’s contextDomain, when the module option is true.
  4. Nothing. useFronticPage alone falls back once more, to the request host.
useFronticClient additionally accepts contextDomain: false to send no domain at all, ignoring the module config:
The resolved value travels as the fs-domain header and is part of every cache key — see Caching.
The context token
A domain resolves to one scope, region and locale — the one you configured. The context token is what a visitor changes when they pick a different region or locale through useFronticContext, and Frontic gives it precedence over the domain. You do not pass it per call. Every composable reads the token from the context cookie and sends it as fs-context, so a switch reaches the next request on its own:
The token is part of every cache key, so the switch moves each active query onto a new entry and its content is fetched rather than reused. Two ways out, per call or per project:
  • contextKey: false on a single call sends no token, so it resolves through the domain alone. Reach for it when the response goes behind a shared cache, where one visitor’s context must not decide what the next one sees.
  • disableContext: true stops the module managing the token at all — nothing is read, nothing is sent, and you pass contextKey yourself.
useFronticContext({ cookieName }) overrides the cookie per instance. The composables resolve their token from the first scope registered that way, since queries have already keyed on it by the time a second one could move it; a second call naming a different cookie warns in development. Pass contextKey explicitly for that second context.

Frontic UI

The module configures Frontic UI — a set of ready-to-use, customizable components designed for Frontic storefronts. When enabled, the module auto-imports your Frontic UI components so they’re available everywhere in your app without manual imports, mirroring Nuxt’s built-in component auto-import.
Frontic UI is currently in open beta. We’d love to hear your feedback! If you have questions or suggestions, please get in touch with us.
string
default:""
Prefix for auto-imported Frontic UI components from your component directory. For example, setting 'Ui' would make Button.vue available as <UiButton />.
string
default:"@/components/ui"
Directory path for Frontic UI components to auto-import.

Composables

boolean | FronticComposable[]
default:"true"
Control which composables are auto-imported during the Nuxt build.
When a composable is disabled, it will not be auto-imported during the Nuxt build. You won’t get type errors or warnings — the composable won’t exist. Use this to reduce bundle size by excluding composables you don’t need.

Composables

The Frontic composables provide a smart data layer for your Nuxt application. Built on Pinia Colada, they handle caching, deduplication, and SSR hydration automatically — instant UI updates via stale-while-revalidate, shared requests across components, and server-to-client data transfer without extra configuration.

useFronticBlock

Fetch single blocks (products, categories) with smart caching

useFronticListing

Fetch listings with automatic caching and SSR support

useFronticSearch

Full-featured search with filters, sorting, and smart state management

useFronticTree

Hierarchical menu trees with automatic caching and SSR support

useFronticPage

Dynamic page routing with redirects and 404 handling

useFronticContext

Locale and region switching with cookie persistence

useFronticClient

Low-level client for direct API access

useFronticBlock

Fetch a single block (product, category, brand) by key with automatic caching and SSR support. See Caching for details.

Parameters

keyof Blocks
required
The name of the block to fetch. Provides full autocomplete for your generated block types.
string | Ref<string>
required
The key identifier for the block. Can be reactive for dynamic fetching.
object
Configuration options.

Returns

Ref<Responses[T] | undefined>
The block data, fully typed based on the block name.
Ref<'pending' | 'error' | 'success'>
Current query status for loading states.
Ref<'idle' | 'loading'>
What the network is doing, independent of what the data is. Together with status this separates a first load (pending) from a background refresh (success while asyncStatus is 'loading') — the state keepPreviousData produces.
Ref<boolean>
Shorthand for asyncStatus === 'loading'.
() => Promise<void>
Refresh data using cache if still valid.
() => Promise<void>
Force a fresh fetch, bypassing cache entirely.

Example


useFronticListing

Fetch a listing by name and parameters with automatic caching and SSR support. Mirrors client.listing(name, params, { query }) — including the optional query for filter, sort, search, limit, and page. See Caching for details.
Building a page with interactive filter/sort/search controls or pagination bound to user input? Use useFronticSearch. useFronticListing is the right pick when the query is fixed at call time — homepage carousels with { limit: 10 }, server-rendered grids with a static filter, etc.

Parameters

keyof Listings
required
The name of the listing to fetch.
ListingParameters[T] | Ref
required
Parameters to pass to the listing endpoint. Type-safe based on listing name.
object
Configuration options.

Returns

Ref<Responses[T] | undefined>
The listing data with items and metadata.
Ref<'pending' | 'error' | 'success'>
Current query status.
Ref<'idle' | 'loading'>
What the network is doing, independent of what the data is. Together with status this separates a first load (pending) from a background refresh (success while asyncStatus is 'loading') — the state keepPreviousData produces.
Ref<boolean>
Shorthand for asyncStatus === 'loading'.
() => Promise<void>
Refresh using cache if valid.
() => Promise<void>
Force fresh fetch.

Example


useFronticSearch

A ready-to-use backend for building stateful search and filter UIs. This composable handles all the interaction logic - filtering, sorting, pagination, and text search - with automatic caching and SSR support, so you can focus on crafting the perfect UI. See Caching for details. It follows best practices to reduce logic overhead in your templates, provides pre-processed filter and sort options with labels and counts, and seamlessly integrates with Frontic UI components.

Parameters

keyof Listings
required
The name of the listing to fetch.
ListingParameters[T] | Ref
required
Parameters to pass to the listing endpoint.
object
Configuration options.

Returns

State & Cache Control
Ref<Responses[T] | undefined>
The full listing response from the API, containing items, pagination metadata, and filter facets.
Ref<'pending' | 'error' | 'success'>
Current query status. Use this to show loading spinners or error states in your UI.
Ref<'idle' | 'loading'>
What the network is doing, independent of what the data is. Together with status this separates a first load (pending) from a background refresh (success while asyncStatus is 'loading') — the state keepPreviousData produces.
Ref<boolean>
Shorthand for asyncStatus === 'loading'.
Ref<SearchState<TFilters, TSorts>>
Pre-processed search state ready for building your UI. Fully typed based on your listing’s filter and sort schema - with IDE autocomplete for filter keys, sort fields, and more. Contains everything needed to render filter sidebars, sort dropdowns, and pagination controls.
Ref<string>
Two-way bindable search term. Connect this directly to your search input - typing automatically triggers debounced API requests when the term exceeds searchTermThreshold.
() => Promise<void>
Re-fetch data, using cache if still valid within staleTime. Use this when you want to ensure fresh data but don’t need to bypass the cache.
() => Promise<void>
Force a fresh fetch, completely bypassing the cache. Use this when you know data has changed and need guaranteed fresh results.
() => Promise<void>
Reset all state (search term, filters, and sorting) to initial values and refresh results. Equivalent to calling resetSearch(), resetFilter(), and resetSorting() together. Use this for a “Clear All” button.
Filter Actions
(field, value) => Promise<void>
Add a single filter value while keeping existing selections. Use this for checkbox-style filters where users can select multiple options.
A { from, to } value is recognised by shape and routed to the range filter list. Ranges do not accumulate: a field holds one range, and a second call replaces it.
(field, value?) => Promise<void>
Remove a specific filter value, or clear all values for a field if no value is provided. Use this when users uncheck options or click “clear” on a filter group.
The field’s range is always cleared, whichever value is passed — a range has no single value to remove.
(field, values) => Promise<void>
Replace all values for a filter field at once. Use this for single-select filters or when setting multiple values programmatically.
When the first entry is a { from, to } object the call is treated as a range and the remaining entries are ignored.
(field?) => Promise<void>
Clear filters for a specific field, or all filters if no field is provided. Use this for “Reset” buttons.
Sort Actions
(sortBy?: SortStringOf<TSorts>) => Promise<void>
Apply a sort order using the 'field:order' format. Call without arguments to reset to default sorting.
() => Promise<void>
Reset to the backend’s default sort order. Equivalent to calling sortResult() without arguments.
Pagination Actions
Filtering, sorting, or searching for a new term returns you to page 1. A pagination component bound to state.active.page follows this on its own; one holding its own page number will drift out of step with the results.
() => Promise<Responses[TSearch] | undefined>
Load the next page. In standard mode (the default), the current page is replaced with the next one. In infinite mode (infinite: true), items from the next page are appended to the existing list — use this for “Load More” buttons or infinite scroll.
() => Promise<Responses[TSearch] | undefined>
Load the previous page. In standard mode, the current page is replaced with the previous one. In infinite mode, items from the previous page are prepended to the existing list.
(page: number) => Promise<Responses[TSearch] | undefined>
Jump to a specific page number. In standard mode, the current page is replaced with the requested page. In infinite mode, all loaded pages are reset and the requested page becomes the new starting point.
Search Actions
Clear the search term and refresh results. Use this for a “clear search” button.

Example

Why use a wrapper composable?
  1. Centralized configuration - All your filter options, sort labels, and search settings live in one file. Need to add a new filter? Update it once.
  2. Shared state - The cacheKey option enables automatic state sharing. When multiple components call the same wrapper, they share the reactive searchTerm, filter/sort state, result data, and pagination.
  3. Clean components - Your page and filter components stay focused on rendering, not configuration.
Any action (like addFilter) called from one component automatically updates all other components using the same cacheKey.

Restoring state from the URL

A search page usually gets its term from the request — ?name=atomic — not from someone typing. Pass it as initialSearch rather than assigning searchTerm:
app/pages/search.vue
Assigning searchTerm does not work for a term that arrives with the request. It commits through a debounced watcher, which never fires before the server render finishes — and after hydration the value has not changed, so the watcher never runs. The page renders unfiltered results and stays that way. searchDebounce: 0 does not help.
The initial* options commit before the first query, so the server renders the right results and the browser does not refetch to correct them. They apply only when the state is empty, so a value restored from the SSR payload — or set by another component sharing the same cacheKey — wins over them. searchTerm remains the right binding for an input the user types into.

useFronticTree

Fetch a Menu Tree — a hierarchical collection of records assembled from a Data Storage and rendered through a Detail Block — with automatic caching and SSR support. See Caching for details.

Parameters

keyof Trees
required
The name of the tree to fetch. Provides full autocomplete for your generated tree types.
object
Configuration options.

Returns

Ref<Responses[T] | undefined>
The full tree response ({ items?: [...] }), fully typed based on the tree name.
ComputedRef<TreeItems<Responses[T]> | undefined>
Shortcut for tree.value?.items. Fully typed based on the tree name — when your generated client is in place, items resolves to the concrete node array, so iterating gives you autocomplete on $items, key, and the block fields defined by the tree’s Detail Block (e.g. name, link).
Ref<'pending' | 'error' | 'success'>
Current query status for loading states.
Ref<'idle' | 'loading'>
What the network is doing, independent of what the data is. Together with status this separates a first load (pending) from a background refresh (success while asyncStatus is 'loading') — the state keepPreviousData produces.
Ref<boolean>
Shorthand for asyncStatus === 'loading'.
() => Promise<void>
Refresh data using cache if still valid.
() => Promise<void>
Force a fresh fetch, bypassing cache entirely.

Example

The cache key includes key, depth, and contextKey, so different subtrees, depth caps, and contexts each cache independently — calling useFronticTree('CategoryNavigation') and useFronticTree('CategoryNavigation', { key: 'shop' }) from different components will not collide.

useFronticPage

Dynamic page routing with automatic slug detection, redirect handling, and 404 errors. Includes automatic caching and SSR support. See Caching for details.

Parameters

string | Ref<string>
The page slug. If omitted, auto-detected from current URL.
object
Configuration options.

Returns

Ref<Page | undefined>
The full page response object.
ComputedRef<Page['data'] | undefined>
The page data payload for rendering.
ComputedRef<string | undefined>
The page type for conditional rendering ('ProductCategory', 'ProductDetail', etc.).
ComputedRef<string | undefined>
The block name to render for this page.
ComputedRef<PageRoute | undefined>
Route information including redirect and context data.
ComputedRef<AlternateRoute[] | undefined>
Alternate language URLs for SEO hreflang tags.
ComputedRef<AlternateRoute | undefined>
Suggested route when context changes (e.g., for locale switching).
Ref<'pending' | 'error' | 'success'>
Current query status.
Ref<'idle' | 'loading'>
What the network is doing, independent of what the data is. Together with status this separates a first load (pending) from a background refresh (success while asyncStatus is 'loading') — the state keepPreviousData produces.
Ref<boolean>
Shorthand for asyncStatus === 'loading'.
() => Promise<void>
Refresh using cache if valid.
() => Promise<void>
Force fresh fetch.

Example

pages/[...slug].vue
The composable automatically constructs the page slug from the current request URL (host + pathname). This works correctly in both SSR and client-side navigation.

useFronticContext

Manage locale and region switching with cookie persistence.

Parameters

object
Configuration options.

Returns

Readonly<Ref<ContextOption[]>>
Available context options with regions and locales.
Readonly<Ref<Context | null>>
Current active context.
Readonly<Ref<string | null>>
The current context token.
(context: { region: string; locale: string }) => Promise<void>
Switch to a different region/locale combination.
Readonly<Ref<boolean>>
Whether the context itself is resolving. Content that depends on it refetches on its own — the token is part of every query key, so a switch re-keys each query — and every composable reports that through its own asyncStatus.
() => Promise<void>
Manually refresh available contexts.

Example


useFronticClient

Low-level client for direct API access. All other composables use this internally.

Parameters

object
Configuration options for client behavior.

Returns

Returns a FronticClient instance with type-safe methods:
<T>(name, key, config?) => Promise<Responses[T]>
Fetch a block by name and key.
<T>(name, params, config?) => Promise<Responses[T]>
Fetch a listing with parameters and query options.
<T>(name, config?) => Promise<Responses[T]>
Fetch a menu tree by name. Pass key and/or depth under config.query to fetch a subtree or cap the depth.
(slug, config?) => Promise<Page>
Fetch a page by its slug.
(token, config?) => Promise<Context>
Get context by token.
(token?, config?) => Promise<[ContextOption[], string]>
Get available contexts.
(context, token, config?) => Promise<Context>
Update context with new region/locale.
(domain, index?, config?) => Promise<string>
Fetch the sitemap XML for a Frontic domain. See Sitemap.

Example


Server routes

useFronticClient is a Nuxt app composable — it reads useNuxtApp, useRequestURL and the app’s runtime config, none of which exist in Nitro. A server/ route uses createFronticServerClient(event) instead:
server/api/stock.get.ts
It returns the same FronticClient the composables use, and resolves from the incoming request what the app-side client resolves from the Nuxt app: Failures are announced on the Nitro frontic:fetch:error hook, with the payload the app-side hook carries — see Error reporting.

Parameters

H3Event
required
The request being handled. Its cookies and URL are what the client resolves from.
object
Overrides for what the event would otherwise decide.
contextDomain: true reads the active @nuxtjs/i18n locale, and there is no i18n instance in Nitro. A project configured that way resolves no domain here rather than guessing one — pass contextDomain per call, or address pages by a slug that already carries the domain.
There is no cache. This is a client, not a query: nothing is deduplicated, nothing is stored, and nothing reaches the SSR payload. For data a page renders, use the composables; for a route’s own response, cache it with Nitro’s own tools.

Hooks

The module fires Nuxt hooks as it works, so you can instrument a storefront without wiring anything per call site. Register them in a plugin.
app/plugins/frontic.ts
Every hook name and payload is typed, so payload.trigger autocompletes and a misspelled hook name is a compile error. frontic:listing:resolved carries a trigger telling you why the listing changed:
That distinction is only available inside the module. From outside you can see that data changed, but not whether a shopper applied a filter or a cache revalidated — so listing analytics built on reactive state count both. Cache revalidations never fire the hook, and one shopper action produces one event however many components read the same listing. Items carry key, an index that is absolute across pages, and the raw row:
app/plugins/frontic.ts
frontic:search:committed fires only once a term stops changing. Instrumenting off the search input directly records s, sk, ski, skis as four searches, which turns your top-queries report into a list of prefixes:
app/pages/search.vue
These knobs are separate from searchDebounce, which controls how fast results update. The two answer different questions and should not share a value.

Error reporting

frontic:fetch:error fires wherever a Frontic call fails, carrying the operation, the identity of what was fetched (block, listing or tree name, page slug, or sitemap domain), the resolved context, the HTTP status, and the error itself. The payload includes a chain — the error’s cause chain, outermost first:
Connection failures arrive as a bare Bad Gateway, with the real reason in cause. Because cause is non-enumerable, it never appears in console.error or in anything that serialises the error — so without the chain, every upstream failure looks identical.
This hook fires on two runtimes. Composable and SSR calls run in the Nuxt app; the proxy route and createFronticServerClient run in Nitro. To catch every failure, register in both.
server/plugins/frontic-errors.ts
Reporting never changes control flow: the error is rethrown untouched, so a failed call still rejects and you can never mistake a failure for an empty result.
The listing, search and page hooks are client-only, so you never double-count on hydration. frontic:fetch:error fires during SSR too — a request that dies while rendering is exactly the failure worth reporting.

Proxy

The module includes a built-in proxy to prevent CORS issues on client-side requests.
1

Browser Request

Client sends request to your server at /api/frontic
2

Server Forwards

Your Nuxt server forwards the request to the Fetch API
3

Response Returns

Response flows back through your server to the browser
Server-side requests (SSR) go directly to the Fetch API without using the proxy.

What the proxy forwards

The proxy passes your request headers through to the Fetch API, minus three it strips first: Every other header is forwarded unchanged, including fs-context, fs-version, fs-domain and fs-request-url. fs-secret is injected server-side when you set fetchApiSecret.
If you run your own proxy route instead of the module’s, strip cookie and authorization yourself. Header forwarding keeps everything except a handful of hop-by-hop names, so a browser-side call otherwise hands your session cookies to the Fetch API.

Timeouts

Calls that go through the proxy are bounded by proxyTimeout, 10 seconds by default. Server-rendered requests reach the Fetch API directly and are not covered by it:
nuxt.config.ts

Sitemap

Frontic generates an XML sitemap for every domain of your project from the pages marked Visible in Sitemap. With sitemap: true, the module serves it from your Nuxt app:
Requests to /sitemap.xml — and /sitemap1.xml, /sitemap2.xml, … when the sitemap is split — are answered with the sitemap for the matching domain, on every locale prefix. https://demo-shop.com/sitemap.xml and https://demo-shop.com/uk/sitemap.xml each serve their domain’s sitemap, ready to submit to search engines or reference from your robots.txt:
robots.txt
The route resolves the domain from the incoming host and path prefix — the same way URLs resolve — injects your fetchApiSecret server-side, and forwards the sitemap’s ETag so crawlers revalidate instead of re-downloading. Responses are cached for ten minutes with stale-while-revalidate. When your config declares the app’s domains (sitemap: { domains: [...] }, a contextDomain string, or contextDomain: true with i18n locales carrying contextDomain properties), the route serves only those domains and answers 404 for any other host. With no declared domains the route serves whatever well-formed host the request carries, and the module warns at build time — declare your domains on any deployment whose forwarded host header is client-controlled.
With @nuxtjs/robots installed, the module adds the Sitemap: directive to your robots.txt automatically. With @nuxtjs/sitemap installed, sitemap: true is ignored with a warning — both would claim /sitemap.xml. An integration that feeds Frontic pages into @nuxtjs/sitemap is planned.
To assemble your own sitemap instead — for example to merge Frontic pages into a larger set — leave the option off and fetch the XML through client.sitemap:

TypeScript

The module configures a path alias so you can import from your generated Frontic client:
This maps @frontic/stack/* to .frontic/* in your project root, where the Frontic CLI generates your typed client.
All composables are fully typed with generics that preserve the specific block/listing types through to the return values, providing full IDE autocomplete for response data.

Caching

All composables use Pinia Colada for intelligent caching:

Stale-While-Revalidate

Shows cached data immediately while fetching fresh data in the background

Automatic Deduplication

Multiple components requesting the same data share a single request

SSR Hydration

Data fetched on server transfers to client without duplicate requests

What a Cache Entry Is Keyed By

requestUrl is deliberately not in the key: it is per-route request metadata, not content identity, and keying on it would split the cache into one entry per visited URL. If you are putting these responses behind a server cache of your own, pass requestUrl: false so the outgoing request stops varying per visitor too — and contextKey: false alongside it, for the same reason. Alongside the composable’s own arguments, every cache key carries the resolved context domain and the active context token. Two locales resolving to different domains therefore never share an entry, and switching context moves every active query onto a new entry rather than reusing the old one — which is what fetches the new context’s content. This matters most when one process serves several domains — prerendering a multi-locale site, warming a cache, or rendering multiple tenants — where a domain-blind key would hand the second request the first one’s payload.

Configure Cache Duration

staleTime decides how long fetched data is trusted; gcTime decides how long an entry nobody is reading is kept before it is dropped.

Data on Screen While a Key Changes

A cache entry is keyed by everything the request depends on, so changing a filter, a page or a context is a different entry — with no data of its own yet. By default the composables hold the previous entry’s data on screen while the new one loads, rather than blanking the view back to pending:
The held value is a placeholder: it is never written to the cache, and the first load of all still reports pending. Turn it off with keepPreviousData: false where the stale content would mislead rather than reassure. useFronticPage already defaults to false, because there a changed key is a different page rather than the same one reloading.

Seeding Data That Arrived With the Page

Frontic lets a page define the data it needs, so a category page’s block can carry its listing in the same payload as the page. initialData hands that payload to the composable instead of fetching it again:
The seed is written to the cache entry, so the query reports success and stays quiet while it is within staleTime. It survives SSR: the entry is serialised into the payload, and the client hydrates it without re-fetching. Which raises the question of how old the payload is. The seed is treated as produced now unless you say otherwise, so a listing that travelled inside a cached page document is trusted as if it had just been fetched — stale prices and stock, marked fresh, for as long as staleTime lasts. Stamp it when the page document can outlive the request that renders it:
Prerendering, ISR and a CDN in front of the page all produce this: one document, many visitors, one timestamp that is not now for any of them.
Seed only data that is the response. A seeded query is indistinguishable from a fetched one — same status, same data — so nothing downstream can tell that the numbers came from somewhere else. To merely avoid a skeleton on data that might be wrong, leave keepPreviousData to do it.
Only the query the composable starts on is seeded. The payload answers one question — one set of filters, one sort, one page — so the first filter, sort or page the user picks fetches for real rather than being handed data that answers a different one.
The query it starts on is not always the plain one. initialFilters, initialSort and any state restored from the URL are applied before the seed is matched, so a filtered start is still a start. Seeding an unfiltered page payload there is accepted, marked fresh, and renders the wrong products — with nothing to distinguish it from a real response.Return undefined whenever the payload does not answer the question the search actually starts with:
In development the composable warns when it starts with filters, a sort, a search term or a page applied and initialData is set.
useFronticSearch takes the same single response in infinite mode and wraps it into the accumulated page set itself:

Manual Cache Control

All composables return two methods for cache control: