Configure Amazon Bedrock Embeddings

This guide shows how to configure Amazon Bedrock embeddings for semantic search in Neo4j Agent Memory.

Supported Models

Neo4j Agent Memory supports these Bedrock embedding models:

Model Dimensions Notes

amazon.titan-embed-text-v2:0

1024

Recommended - Best quality/performance balance

amazon.titan-embed-text-v1

1536

Legacy model, higher dimensionality

cohere.embed-english-v3

1024

Optimized for English text

cohere.embed-multilingual-v3

1024

Multilingual support (100+ languages)

Prerequisites

  • AWS account with Bedrock access

  • Model access enabled in AWS Console (Bedrock → Model access)

  • boto3 installed: pip install neo4j-agent-memory[bedrock]

Basic Configuration

Using BedrockEmbedder Directly

from neo4j_agent_memory.embeddings import BedrockEmbedder

embedder = BedrockEmbedder(
    model="amazon.titan-embed-text-v2:0",
    region_name="us-east-1",
)

# Generate a single embedding
embedding = await embedder.embed("Hello, world!")
print(f"Dimensions: {len(embedding)}")  # 1024

# Check model dimensions
print(f"Model dimensions: {embedder.dimensions}")

Using with MemoryClient

from neo4j_agent_memory import MemoryClient, MemorySettings
from neo4j_agent_memory.config.settings import (
    Neo4jConfig,
    EmbeddingConfig,
    EmbeddingProvider,
)

settings = MemorySettings(
    neo4j=Neo4jConfig(
        uri="neo4j+s://xxx.databases.neo4j.io",
        password="your-password",
    ),
    embedding=EmbeddingConfig(
        provider=EmbeddingProvider.BEDROCK,
        model="amazon.titan-embed-text-v2:0",
        aws_region="us-east-1",
    ),
)

async with MemoryClient(settings) as client:
    # Embeddings are generated automatically
    await client.short_term.add_message(
        session_id="session-1",
        role="user",
        content="Hello!",
        generate_embedding=True,
    )

Using with Environment Variables

export EMBEDDING_PROVIDER=bedrock
export EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
export AWS_REGION=us-east-1
from neo4j_agent_memory import MemoryClient, MemorySettings

# Settings automatically read from environment
settings = MemorySettings.from_env()

Authentication

BedrockEmbedder uses the standard AWS credential chain:

  1. Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY)

  2. Shared credentials file (~/.aws/credentials)

  3. IAM role (for EC2, Lambda, ECS)

Using a Named Profile

embedder = BedrockEmbedder(
    model="amazon.titan-embed-text-v2:0",
    profile_name="my-profile",
)

Using Explicit Credentials

embedder = BedrockEmbedder(
    model="amazon.titan-embed-text-v2:0",
    aws_access_key_id="AKIA...",
    aws_secret_access_key="...",
    region_name="us-east-1",
)

Batch Processing

BedrockEmbedder supports efficient batch embedding with automatic chunking:

texts = [
    "First document about machine learning",
    "Second document about databases",
    "Third document about cloud computing",
]

# Automatically batches in groups of 25
embeddings = await embedder.embed_batch(texts)
print(f"Generated {len(embeddings)} embeddings")

Bedrock limits batch size to 25 texts per request. The embedder automatically splits larger batches and processes them sequentially.

Configuration Options

Option Default Description

model

amazon.titan-embed-text-v2:0

Bedrock model ID

region_name

From env/config

AWS region for Bedrock API

profile_name

None

AWS credentials profile name

batch_size

25

Max texts per batch request

IAM Permissions

Required IAM policy for Bedrock embedding access:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "bedrock:InvokeModel"
            ],
            "Resource": [
                "arn:aws:bedrock:*::foundation-model/amazon.titan-embed-*",
                "arn:aws:bedrock:*::foundation-model/cohere.embed-*"
            ]
        }
    ]
}

For Lambda functions, attach this policy to the execution role.

Choosing a Model

Titan Text V2 (Recommended)

Best for most use cases:

  • 1024 dimensions - efficient storage in Neo4j

  • Strong semantic understanding

  • Good performance/cost balance

embedder = BedrockEmbedder(model="amazon.titan-embed-text-v2:0")

Cohere Multilingual

Best for international content:

  • Supports 100+ languages

  • Same 1024 dimensions as Titan V2

  • Better cross-lingual search

embedder = BedrockEmbedder(model="cohere.embed-multilingual-v3")

Troubleshooting

AccessDeniedException

Ensure model access is enabled:

  1. Go to AWS Console → Bedrock → Model access

  2. Click "Manage model access"

  3. Enable the embedding models you need

  4. Wait for access to be granted (may take a few minutes)

ValidationException

Check your model ID format:

  • ✅ Correct: amazon.titan-embed-text-v2:0

  • ❌ Incorrect: titan-embed-text-v2 (missing prefix)

Region Not Available

Bedrock is not available in all regions. Supported regions:

  • us-east-1 (N. Virginia) - Most models

  • us-west-2 (Oregon)

  • eu-west-1 (Ireland)

  • ap-northeast-1 (Tokyo)

Slow Performance

For high-volume embedding:

  1. Use embed_batch() instead of multiple embed() calls

  2. Consider running in the same AWS region as your Neo4j database

  3. For Lambda, increase memory allocation (embedding is CPU-bound)

See Also