strands.storage.embeddings.bedrock

Amazon Bedrock embedder for vector search strategies.

BedrockEmbedder

class BedrockEmbedder(Embedder)

Defined in: src/strands/storage/embeddings/bedrock.py:18

An :class:Embedder backed by Amazon Titan Text Embeddings on Amazon Bedrock.

Sends \{"inputText": text} via InvokeModel and returns the embedding field of the response.

Example:

from strands.storage import S3Storage
from strands.storage.embeddings.bedrock import BedrockEmbedder
from strands.storage.search import S3VectorSearchStrategy

storage = S3Storage(
    "my-bucket",
    search_strategy=S3VectorSearchStrategy(
        vector_bucket_name="my-vectors",
        index_name="notes",
        embedder=BedrockEmbedder(model_id="amazon.titan-embed-text-v2:0", region_name="us-west-2"),
    ),
)

__init__

def __init__(*,
             model_id: str = DEFAULT_EMBEDDER_MODEL_ID,
             region_name: str | None = None,
             boto_session: boto3.Session | None = None,
             boto_client_config: BotocoreConfig | None = None) -> None

Defined in: src/strands/storage/embeddings/bedrock.py:41

Initialize the embedder.

Arguments:

• model_id - Bedrock model ID or ARN of an Amazon Titan Text Embeddings model. Other model families use a different request and response shape and are not supported yet; subclass :class:Embedder for those.
• region_name - AWS region override for the Bedrock Runtime client.
• boto_session - Pre-configured boto3 session. Cannot be combined with region_name.
• boto_client_config - Botocore Config object for the Bedrock Runtime client.

Raises:

• StorageError - If both region_name and boto_session are provided.

embed

async def embed(text: str) -> list[float]

Defined in: src/strands/storage/embeddings/bedrock.py:75

Embed text with the configured Titan model.

Arguments:

• text - The text to embed.

Returns:

The embedding vector.

Raises:

• StorageError - If the model response has no embedding list.