Docs
API Documentation
Searchkit API

Installation

It is available as searchkit on npm.

npm install searchkit

Then import it in your project:

import Searchkit from "searchkit";
// OR if you are using CDN
const Searchkit = window.Searchkit;

Usage

You can use Searchkit with either the @searchkit/api or the @searchkit/instantsearch-client package.

 
import Client from '@searchkit/instantsearch-client'
import Searchkit, { SearchkitConfig } from "searchkit"
 
const searchkitClient = new Searchkit({
  connection: {
    host: "https://commerce-demo.es.us-east4.gcp.elastic-cloud.com:9243",
    // cloud_id: "my-cloud-id" if using Elastic Cloud
 
    apiKey: "a2Rha1VJTUJMcGU4ajA3Tm9fZ0Y6MjAzX2pLbURTXy1hNm9SUGZGRlhJdw==", // optional apiKey
    headers: { // optional headers sent to Elasticsearch or elasticsearch proxy
      "my-custom-header": "my-custom-value"
    },
    auth: {
      username: 'elastic',
      password: 'changeme'
    }
  },
  search_settings: {
    highlight_attributes: ["title", "actors"],
    search_attributes: ["title", "actors"],
    result_attributes: ["title", "actors", "poster", "year"],
    facet_attributes: [
      { attribute: "type", field: "type", type: "string" },
      { attribute: "actors", field: "actors.keyword", type: "string" },
      { attribute: "rated", field: "rated", type: "string" },
      { attribute: "imdbrating", type: "numeric" },
      { attribute: "metascore", type: "numeric" },
    ]
  },
})
 
const searchClient = Client(searchkitClient);
 
export const App = () => {
  return (
    <InstantSearch searchClient={searchClient} indexName="movies">
      <SearchBox />
      <Hits />
    </InstantSearch>
  )
}

SearchkitConfig

connection configuration

  • host - Elasticsearch url host
  • cloud_id - Elasticsearch cloud id. This is optional but recommended way to connect to Elasticsearch on Cloud. You can find your cloud id in the Elastic Cloud console (opens in a new tab).
  • apiKey - Elasticsearch API key. This is optional but strongly recommended for production environments. You can create an API key via Kibana (opens in a new tab).
  • headers - Additional headers to pass to Elasticsearch. This is optional.
  • withCredentials - Whether to send credentials with the request. This is optional. Useful for CORS requests. Default is false.
  • auth - Basic auth credentials. This is optional.

See connecting to Elasticsearch for more information on setting up Elasticsearch.

Custom Transporter

Connection also supports implementing your own network transporter to Elasticsearch. You can do this by passing in a transporter instance through connection.

You can use this for more complex authentication connections with Elasticsearch or Opensearch.

 
import { ESTransporter } from 'searchkit'
import type { SearchRequest } from "searchkit"
 
class MyTransporter extends ESTransporter {
  async performNetworkRequest(requests: SearchRequest[]) {
    // you can use any http client here
    return fetch(`https://localhost:9200/_msearch`, {
      headers: {
        // Add custom headers here
      },
      body: this.createElasticsearchQueryFromRequest(requests),
      method: 'POST'
    })
  }
}
 
// then pass the custom transporter to the client
const client = Client({
  connection: new MyTransporter()
});

search_settings configuration

search_settings: {
  search_attributes: ["title^3", "actors"],
  result_attributes: ["title", "actors", "poster", "year", "rating"],
  facet_attributes: [
    { attribute: "type", field: "type", type: "string" },
    { attribute: "actors", field: "actors.keyword", type: "string" },
    { attribute: "rated", field: "rated", type: "string" },
    { attribute: "imdbrating", type: "numeric" },
    { attribute: "metascore", type: "numeric" },
  ],
  filter_attributes: [
    { attribute: "year", field: 'year', type: "numeric" }
  ],
  highlight_attributes: ["title", "actors"],
  snippet_attributes: [
    "description",
    "plot:200"
  ],
  sorting: {
    default: {
      field: '_score',
      order: 'desc'
    },
    _year_desc: [{
      field: 'year',
      order: 'desc'
    }]
  },
  geo_attribute: "location",
  runtime_mappings: {
    rating: {
      type: 'keyword',
      script: {
        source: "emit(doc['rated'].size()>0 ? doc['rated'].value : '')"
      }
    }
  },
},

Attributes that are used to configure the search experience.

  • search_attributes - Attributes that are used to search the results.
  • result_attributes - Attributes that are returned in the search result response.
  • facet_attributes - Attributes that are used to create facets. Facets can be of type string or numeric or date.
  • filter_attributes - Attributes that are used to create filters. Filters can be of type string or numeric or date.
  • highlight_attributes - Attributes that are used to highlight the search results.
  • snippet_attributes - Attributes that are used for highlighting search results for long fields.
  • sorting - Attributes that are used to create sorting options. Sorting can be a single sorting field or multiple.
  • query_rules - Rules that affect search relevance. See Query Rules for more information.
  • geo_attribute - Attribute that is used for geo based searches.
  • runtime_mappings - Runtime mappings that are used to transform fields in the search results. See Runtime Mappings for more information. Once used, you are able to use the transformed field in the search_attributes, result_attributes, facet_attributes and filter_attributes configuration.
search_attributes

The search attributes define what Elasticsearch fields should be searched when a user performs a search.

The search attributes can be configured as follows:

search_attributes: [
  "description", 
  "actors", 
  { field: "title", weight: 3 }, 
  "released.year"
];

The following configuration will search the description, actors and an object field released.year fields with the default weight of 1. The title field will be weighted 3 times more than the actors field.

facet_attributes

For text based facets, you need to specify a keyword type field to use for the facet. This is because Elasticsearch does not support aggregations on text fields. You can find more information about the field mapping here (opens in a new tab).

Typically you would create a subfield of the text field with the keyword type. For example, if you have a actors field with the text type, you would create a actors.keyword field with the keyword type.

{
  "mappings": {
    "properties": {
      "actors": {
        "type": "text",
        "fields": {
          "keyword": {
            "type": "keyword",
            "ignore_above": 256
          }
        }
      }
    }
  }
}

in the above example, you would specify the actors facet as follows:

facet_attributes: [
  { attribute: "actors", field: "actors.keyword", type: 'string' },
],

and you would use the actors attribute within the UI components.

Below is an example of a RefinementList Instantsearch React component that uses the actors facet.

<RefinementList attribute="actors" searchable={true} limit={10} />

See facets guide for more information.

Custom Aggregations & Filter Queries

Searchkit uses the terms aggregation to generate facets and applies filters with term clause. You can use the facetQuery & filterQuery option to specify a custom aggregation query.

facet_attributes: [
  {
    field: 'actors.keyword',
    attribute: 'actors',
    type: 'string',
    // Custom aggregation query
    facetQuery: () => ({
      filters: {
        filters: {
          movie: {
            term: {
              type: 'movie'
            }
          },
          episode: {
            term: {
              type: 'episode'
            }
          }
        }
      }
    }),
    // handle the aggregation response and return an object with the facet names and values.
    // here we are hardcoding the values to 100 for each filter bucket
    facetResponse: (aggregation: AggregationsFiltersAggregate) => {
      const buckets = aggregation.buckets as AggregationsFiltersBucket[]
      return Object.keys(buckets).reduce(
        (sum, bucket) => ({
          ...sum,
          [bucket]: 100
        }),
        {}
      )
    },
    // When a user selects a facet, this function is called to generate the filter query
    filterQuery: (field: string, value: string) => {
      return { match: { ['type.keyword']: value } }
    }
  }
],
numeric and date types

For range based facets, you need to specify the type as numeric or date. This is so the client can correctly generate facet stats for the field to be used in the UI.

facet_attributes: [
  { attribute: "imdbrating", field: "imdbrating", type: "numeric" },
  { attribute: "metascore", field: "metascore", type: "numeric" },
],

Below is an example of a NumericMenu Instantsearch React component that uses the imdbrating facet.

<NumericMenu
  attribute="imdbrating"
  items={[
    { label: "5 - 7", start: 5, end: 7 },
    { label: "7 - 9", start: 7, end: 9 },
    { label: ">= 9", start: 9 },
  ]}
/>
filter_attributes

Similar to facet_attributes, filter_attributes are used to apply for filters. The difference is that filter_attributes are not used to generate facets.

filter_attributes: [
  { attribute: "writers", field: 'writers', type: "string" }
],
Custom Filter Queries

Searchkit uses the term clause to apply filters. You can use the filterQuery option to specify a custom filter query.

filter_attributes: [
  {
    field: 'actors.keyword',
    attribute: 'actors',
    type: 'string',
    filterQuery: (
      field, // 'actors.keyword' in this case
      value // value specified by the frontend
    ) => ({
      match: { // example of using match clause instead of terms filter
        [field]: value
      }
    }
    )
  }
],
sorting

Sorting can be configured as follows:

 
sorting: {
  default: {
    field: '_score',
    order: 'desc'
  },
  _year_desc: [{
    field: 'year',
    order: 'desc'
  }]
}

The default sorting option is used when no sorting option is selected. The _year_desc sorting option is used when the user selects the _year_desc sorting option.

The sorting option can be a single sorting field or multiple sorting fields.

See the sorting guide for more information.

highlight_attributes

The highlight_fields is used to configure the fields that are highlighted in the search results. The highlight_fields must point to a text field type in Elasticsearch.

highlight_attributes: ["title"]

Use with the Highlight component to display the highlighted fields.

snippet_attributes

The snippet_attributes is used to configure the attributes that are highlighted in the search results. The snippet_attributes must point to a text field type in Elasticsearch.

`snippet_attributes: [
  'long_bio',
  'description:200'
]
geo_attribute

The geo_attribute is used to configure the geo based search experience. The geo_attribute must point to a geo_point or geo_shape field type in Elasticsearch.

geo_attribute: "location"

See Geo Search for more information.

Request Options

RequestOptions is an object that contains the following properties:

  • getQuery - The function used to override the default Elasticsearch query.
  • getBaseFilters - The function used to provide the base Elasticsearch filters. These filters are applied to all search requests.
  • hooks - The hooks object containing the following properties:
    • beforeSearch - The function that is invoked before the search request is executed.
    • afterSearch - The function that is invoked after the search response is received.

getQuery optional function

The getQuery function is used to override the default Elasticsearch query. The function must return an Elasticsearch query. You can read more about the Elasticsearch query DSL here (opens in a new tab).

Below is an example of a getQuery function that overrides the default query to use the combined_fields query type (read more about combined_fields here (opens in a new tab)).

To see the full Elasticsearch query that is executed to Elasticsearch, you can run the client in debug mode (see below).

  const results = await client.handleRequest(req.body, {
    getQuery: (query, search_attributes) => {
      return [
        {
          combined_fields: {
            query,
            fields: search_attributes,
          },
        },
      ];
    }
  });
Example: Exclude BM25 Query

If you're using KNN exclusively, you might want to exclude the BM25 query from the Elasticsearch query. You can do this by overriding the getQuery function and returning false.

  const results = await client.handleRequest(req.body, {
    getQuery: () => {
      return false
    }
  });

getKnnQuery optional function

If you want to specify a KNN query, you can use the getKnnQuery function. The function must return an Elasticsearch KNN query. You can read more about the KNN query DSL here (opens in a new tab).

  const results = await client.handleRequest(req.body, {
    getKnnQuery(query, search_attributes, config) {
      return {
        field: 'dense-vector-field',
        k: 10,
        num_candidates: 100,
        query_vector_builder: {
          text_embedding: {
            model_id: 'cookie_model',
            model_text: query
          }
        }
      }
    },
    // Optional: You may want to exclude the BM25 query
    getQuery: () => {
      return false
    }
  });
Function Parameters
  • query: The query string from the search request.
  • search_attributes: The search attributes from the search configuration.

getBaseFilters optional function

The getBaseFilters function is used to add filters to the Elasticsearch query. The function must return an Elasticsearch query. You can read more about the Elasticsearch query DSL here (opens in a new tab).

This function is useful if the request needs to be filtered based on the user's session. For example, if you want to filter the search results based on the user's role or status.

Below is an example of a getBaseFilters function that adds a filter to the Elasticsearch query to only return results where the status field is published.

To see the full Elasticsearch query that is executed to Elasticsearch, you can run the client in debug mode (see below).

  const results = await client.handleRequest(req.body, {
    getBaseFilters: () => {
      return [
        {
          bool: {
            must: {
              term: {
                status: {
                  value: "published",
                },
              },
            },
          },
        },
      ];
    }
  });

RequestOptions Hooks

Hooks are functions that are invoked at different stages of the search request. Hooks are useful if you want to perform some action before or after the search request is executed.

beforeSearch hook function

The beforeSearch hook is invoked before the search request is executed. This hook is useful if you want to perform some action before the search request is executed.

Examples include:

  • Learn to rank
  • Semantic search
  • A/B testing

Below is an example of a beforeSearch hook that adds a track_total_hits to the Elasticsearch query.

  client = SearchkitInstantsearchClient(sk, {
    hooks: {
        beforeSearch: async (searchRequests) => {
 
            return searchRequests.map((sr) => {
                return {
                  ...sr,
                  body: {
                    ...sr.body,
                    track_total_hits: true
                 }
              }
           })
 
        }
    }
})

To see the full Elasticsearch query that is executed to Elasticsearch, you can run the client in debug mode (see below).

Function Parameters
  • searchRequests - An array of SearchRequest objects. Each SearchRequest object contains the following properties:
    • indexName - The Elasticsearch index name.
    • body - The Elasticsearch request body query.
    • request - The state requested from the UI. Contains attributes like query, filters, sort, size and more.

afterSearch hook function

The afterSearch hook is invoked after the search response is received. This hook is useful if you want to perform some action after the search response is received.

Examples include:

  • Logging
  • Analytics

Below is an example of an afterSearch hook that logs the search response to the console.

  const results = await client.handleRequest(req.body, {
    hooks: {
      afterSearch: (searchRequests, searchResponses) => {
        console.log(searchResponses);
        return searchResponses;
      },
    },
  });

Debug mode

The client can be run in debug mode to help with debugging the Elasticsearch query. To run the client in debug mode, set the debug flag to true in the Client function.

This is helpful when you're overriding the query via getQuery or providing base filters via getBaseFilters and want to see the Elasticsearch query that is executed to Elasticsearch.

const client = Client({
  // search_settings configuration
  connection: {
    // ...
  },
  search_settings: {
    search_attributes: ["title", "plot"],
    // ...
  }
}, { debug: true });

When the client is run in debug mode, the Elasticsearch query is logged to the console.

Meta fields

Score: Accessible in the search results hit as _score. This is the relevance score of the document. Index: Accessible in the search results hit as _index. This is the Elasticsearch index name.

FAQ

How do I search across multiple indices?

You can search across multiple indices by specifying the index names in the indexName parameter as comma-separated values.

You can also create an elasticsearch alias that points to multiple indices and use the alias name in the indexName parameter.

export const App = () => {
  return (
    <InstantSearch searchClient={searchClient} indexName="movies,episodes,series">
      <SearchBox />
      <Hits />
    </InstantSearch>
  )
}

Apache 2.0 2024 © Joseph McElroy.
Need help? Join discord