Semantic Search with Cohere Embeddings

Set up Meilisearch with Cohere embedding models (v3 and v4) for semantic search.

Cohere provides high-quality multilingual embedding models. This guide shows you how to configure Meilisearch with Cohere embeddings using the REST embedder, covering both the v3 and the newer v4 model families.

Requirements

• A Meilisearch project
• A Cohere account with an API key

Available models

Embed v4 (recommended)

Model	Dimensions
embed-v4.0	1536

Embed v4 is Cohere’s latest model with improved quality and multilingual support. It uses the v2 API endpoint.

Embed v3

Model	Dimensions
embed-english-v3.0	1024
embed-multilingual-v3.0	1024
embed-english-light-v3.0	384
embed-multilingual-light-v3.0	384

Configure the embedder

Embed v4 (v2 API)

{
  "cohere": {
    "source": "rest",
    "apiKey": "<COHERE_API_KEY>",
    "dimensions": 1536,
    "documentTemplate": "A product named '{{doc.name}}': {{doc.description}}",
    "url": "https://api.cohere.com/v2/embed",
    "request": {
      "model": "embed-v4.0",
      "texts": ["{{text}}", "{{..}}"],
      "input_type": "search_document",
      "embedding_types": ["float"]
    },
    "response": {
      "embeddings": {
        "float": ["{{embedding}}", "{{..}}"]
      }
    }
  }
}

Embed v3 (v1 API)

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

The v4 and v3 models use different API endpoints (/v2/embed vs /v1/embed) and different response formats. Make sure the url and response fields match the model family you choose.

Replace <COHERE_API_KEY> with your actual Cohere API key. Adjust dimensions and model to match the model you select.

Send this configuration to Meilisearch by updating your index settings:

curl \
  -X PATCH 'MEILISEARCH_URL/indexes/INDEX_NAME/settings' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer MEILISEARCH_KEY' \
  --data-binary '{
    "embedders": {
      "cohere": {
        "source": "rest",
        "apiKey": "<COHERE_API_KEY>",
        "dimensions": 1536,
        "documentTemplate": "A product named '\''{{doc.name}}'\'': {{doc.description}}",
        "url": "https://api.cohere.com/v2/embed",
        "request": {
          "model": "embed-v4.0",
          "texts": ["{{text}}", "{{..}}"],
          "input_type": "search_document",
          "embedding_types": ["float"]
        },
        "response": {
          "embeddings": {
            "float": ["{{embedding}}", "{{..}}"]
          }
        }
      }
    }
  }'

Meilisearch handles batching and rate limiting automatically. Monitor the tasks queue to track indexing progress.

Test the search

{
  "q": "comfortable shoes for walking",
  "hybrid": {
    "semanticRatio": 0.5,
    "embedder": "cohere"
  }
}

Next steps

• Document template best practices to optimize which fields are embedded
• Custom hybrid ranking to tune the balance between keyword and semantic results
• Embedder settings reference for all configuration options