Cohere
Parameter reference for CohereEmbedder: model ID, input type, embedding types, and batching options.
Cohere Embedder embeds documents using Cohere's embedding models.
dimensions is the vector width Agno expects. It is not sent to Cohere and does not change the model's output width. The default embed-english-v3.0 model returns 1024 dimensions, while CohereEmbedder inherits a default of 1536. Set dimensions=1024 when using the default model.
Parameters
| Parameter | Type | Description | Default |
|---|---|---|---|
id | str | The model ID to use for embeddings | "embed-english-v3.0" |
dimensions | Optional[int] | Vector width expected by Agno and the vector database. This value is not sent to Cohere. | 1536 |
input_type | str | Type of input being embedded (e.g., "search_query", "classification", "clustering") | "search_query" |
embedding_types | Optional[List[str]] | Cohere embedding response types. Include "float" when setting this option because Agno reads float embeddings. | None |
api_key | Optional[str] | Cohere API key. When omitted, the Cohere client reads CO_API_KEY. | None |
request_params | Optional[Dict[str, Any]] | Additional parameters for embedding requests | None |
client_params | Optional[Dict[str, Any]] | Additional parameters for client initialization | None |
cohere_client | Optional[CohereClient] | Pre-configured Cohere client | None |
async_client | Optional[AsyncCohereClient] | Pre-configured async Cohere client | None |
exponential_backoff | bool | Retry Cohere rate-limit responses during asynchronous batch embedding. This applies only to the async batch path. | False |
enable_batch | bool | Embed multiple texts per API call when adding content to vector databases | False |
batch_size | int | Number of texts per batch when enable_batch is True | 100 |
Set embedding_types=["float"] when selecting Cohere response types. Agno reads response.embeddings.float_; configurations limited to int8, uint8, binary, or ubinary return an empty embedding.