Skip to main content
Public search finds published records straight in the search engine, with a search token instead of an API key: results come back as fast as you type, and the searches don’t count toward your API requests. Use it from your site’s browser code or from your own backend. Search runs on Meilisearch, so its clients and InstantSearch work as they are. Your servers can also search with an API key, through the API: that returns full records, with expand. You need three things: the search URL, a token, and the index name of each collection you search. Tokens find only collections included in public search. In the web app, open a collection’s Search tab and turn on Include in public search, or turn it on when you create the collection. It is off by default, and it applies to every token of the project, so a collection you include later is searchable with the tokens you already have. Through the API, it is the collection’s public_search. A token finds only the published records of included collections. The search engine enforces that: other collections, other projects and unpublished records stay out of reach, whatever a request asks for. Including or removing a collection reaches every token within seconds.
A token is public: in a site, it goes in your page source. Every field of the published records it finds can be read with it, so keep private content out of public search.

​
Index names

Each collection has its own index. Its name shows on the collection’s Search tab, under Public search index name, while the collection is included in public search. Through the API, it is the index of the collection’s search index:
Index names are opaque and never change, so you can keep them in your code or configuration.

​
Create a token

In the web app, open your project’s Search tokens page, choose New search token and name it after the site or app that will use it. Copy the token from the list. Make one per site or app, so deleting one stops only that site’s searches. A project can have 25 tokens. Send searches to https://search.nohead.io, the search URL for every token (a token’s search_url in the API). It lets searches through and nothing else, and limits how many requests one address can make:
Hits look like records:
  • id (rec_…), and published_at, created_at and updated_at as Unix seconds; published_at can be filtered and sorted on, such as filter: "published_at > 1767225600" for posts published since 2026
  • field values under data, by field key: dates as Unix seconds, rich text as plain text, JSON as a string, assets and relations as IDs
Several collections at once take one query each, or a federated search across them (federation: {} in the request).

​
Filter and sort

Fields take part in searches by their search settings, as in every search: words match searchable fields, and filterable and sortable fields can be used in filters and sorts. Name them under data:
Dates are Unix seconds here too: 'data.starts_on >= 1790812800' finds records from 1 October 2026 (00:00 UTC) on.

​
Facets

Facets count the hits per value of a filterable field, for filter lists such as “French (12)”. Ask for them with facets:
Each result then has a facetDistribution, counted over the hits that match the query and filters, and a facetStats with the lowest and highest value of number and date fields, for range sliders:
  • Fields that facet well have a few values shared by many records: enums, booleans, and short text such as a brand or a city. Fields with several values count each value once per record, so tags work as facets.
  • Relations and assets count by ID (rec_…, ast_…). To show names, look them up, or keep an enum or text field next to the relation.
  • Up to 100 values are returned per facet, and only for published records, like hits.
  • In a federated search, name the facets per index in the federation instead, federation: { facetsByIndex: { [index]: ["data.cuisine"] } }, and read them from the response’s facetsByIndex. A query’s own facets is refused there.
To find values within a facet as someone types (InstantSearch’s refinementList with searchable: true), search them:

​
InstantSearch

​
Rename or delete a token

Rename a token or delete it from its menu on the Search tokens page. A renamed token stays the same. A deleted one fails at once, and your other tokens keep working. To stop searches of one collection, remove it from public search instead. While a project or organization is scheduled for deletion its tokens stop working, and they work again if it is restored.