Configure Cohere embedder

Set up the Cohere embedder for semantic and hybrid search using Cohere’s embedding models.

The Cohere embedder connects Meilisearch to Cohere’s embedding API. Cohere models support multiple languages and offer different model sizes for different performance needs. Since Meilisearch does not have a built-in Cohere source, you configure it using the rest embedder source.

This guide requires a Cohere account with an API key.

Choose a model

Cohere offers several embedding models:

Model	Dimensions	Notes
embed-english-v3.0	1,024	Best accuracy for English content
embed-multilingual-v3.0	1,024	Best for multilingual datasets
embed-english-light-v3.0	384	Faster, lower cost for English content
embed-multilingual-light-v3.0	384	Faster, lower cost for multilingual content

If your dataset is primarily English, use embed-english-v3.0. For multilingual content, choose embed-multilingual-v3.0. The light variants are faster and cheaper but may return slightly less accurate results.

Configure the embedder

Because Cohere uses the REST embedder source, you must define the request and response structures that match Cohere’s API. Create the following embedder configuration:

{
  "my-cohere": {
    "source": "rest",
    "url": "https://api.cohere.com/v1/embed",
    "apiKey": "COHERE_API_KEY",
    "dimensions": 1024,
    "documentTemplate": "A product named '{{doc.name}}' described as '{{doc.description}}'",
    "request": {
      "model": "embed-english-v3.0",
      "texts": ["{{text}}", "{{..}}"],
      "input_type": "search_document"
    },
    "response": {
      "embeddings": ["{{embedding}}", "{{..}}"]
    }
  }
}

In this configuration:

• source: must be "rest" because Cohere uses the REST embedder integration
• url: the Cohere embeddings API endpoint
• apiKey: your Cohere API key
• dimensions: the number of dimensions for the chosen model (1024 for embed-english-v3.0)
• documentTemplate: a Liquid template that converts your documents into text for embedding
• request: defines the structure of requests sent to Cohere, including the model and input format
• response: tells Meilisearch where to find the embeddings in Cohere’s response

The input_type parameter is required by Cohere’s API. Set it to "search_document" when indexing documents. Meilisearch automatically uses "search_query" for search queries.

Update your index settings

Send the embedder configuration to Meilisearch:

curl \
  -X PATCH 'MEILISEARCH_URL/indexes/INDEX_NAME/settings' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer MEILISEARCH_KEY' \
  --data-binary '{
    "embedders": {
      "my-cohere": {
        "source": "rest",
        "url": "https://api.cohere.com/v1/embed",
        "apiKey": "COHERE_API_KEY",
        "dimensions": 1024,
        "documentTemplate": "A product named '\''{{doc.name}}'\'' described as '\''{{doc.description}}'\''",
        "request": {
          "model": "embed-english-v3.0",
          "texts": ["{{text}}", "{{..}}"],
          "input_type": "search_document"
        },
        "response": {
          "embeddings": ["{{embedding}}", "{{..}}"]
        }
      }
    }
  }'

Replace MEILISEARCH_URL with the address of your Meilisearch project, INDEX_NAME with your index name, MEILISEARCH_KEY with your Meilisearch API key, and COHERE_API_KEY with your Cohere API key.

Meilisearch will start generating embeddings for all documents in the index. Monitor progress through the task queue.

Never share your Cohere API key publicly or commit it to version control. Use environment variables or a secrets manager to store it securely.

Test the embedder

Once indexing is complete, perform a search using the hybrid parameter:

{
  "q": "something to stir soup with",
  "hybrid": {
    "semanticRatio": 0.5,
    "embedder": "my-cohere"
  }
}

A semanticRatio of 0.5 returns a balanced mix of keyword and semantic results. Adjust this value based on your needs.

Next steps

Full Cohere guide

In-depth guide with advanced configuration options

Choose an embedder

Compare Cohere with other embedder providers