SYSTEM Cited by 2 sources

Amazon Titan Text Embeddings¶

Amazon Titan Text Embeddings is AWS's family of first-party text embedding models available through Amazon Bedrock. Invoked via the Bedrock Runtime API (bedrock.invoke_model with a modelId like amazon.titan-embed-text-v2:0), it turns a string into a dense float32 vector suitable for similarity search, clustering, RAG, and classification.

This page is a stub created for the cross-reference from the S3 Vectors launch.

(Source: sources/2025-07-16-aws-amazon-s3-vectors-preview-launch)

Usage pattern (from the S3 Vectors launch)¶

import boto3, json
bedrock = boto3.client("bedrock-runtime", region_name="us-west-2")

for text in texts:
    body = json.dumps({"inputText": text})
    response = bedrock.invoke_model(
        modelId="amazon.titan-embed-text-v2:0",
        body=body,
    )
    embedding = json.loads(response["body"].read())["embedding"]

The resulting embedding is inserted into an S3 Vectors index via s3vectors.put_vectors — the paved path for producing vectors to store in S3 Vectors or any other vector store.

Role in the Bedrock RAG stack¶

Embedding model of choice for Bedrock Knowledge Bases at launch-time for customers who don't bring their own.
Distance metric recommended by the model (cosine) should be used when creating the vector index — see concepts/vector-similarity-search.

Caveats¶

This page covers only what the 2025-07-16 S3 Vectors launch surfaces. Exact dimensionality, token limits, and V1-vs-V2 differences are not documented here.

Seen in¶

sources/2025-07-16-aws-amazon-s3-vectors-preview-launch — the worked Python example explicitly calls amazon.titan-embed-text-v2:0 as the paved-path embedding model for populating an S3 vector index.
sources/2025-12-11-aws-architecting-conversational-observability-for-cloud-applications — Titan Embeddings v2 embeds operational telemetry (logs, events, metrics) at ingest time + the user's natural-language query at retrieval time; canonical example of Titan used for telemetry-to-RAG rather than document-RAG.