AWS Bedrock Models¶

The Redis Agent Memory Server supports Amazon Bedrock for both embedding models and LLM generation models. This allows you to use AWS-native AI models while keeping your data within the AWS ecosystem.

See also: LLM Providers for a broader overview of all supported providers, and Embedding Providers for embedding-specific configuration.

Quick Start — Run a Full Bedrock-Backed Instance¶

Follow these steps to get the memory server running entirely on AWS Bedrock in under five minutes.

1. Install the [aws] extra¶

The server's core install does not include AWS SDK libraries. You must install the [aws] extra so that boto3 (AWS SDK for Python) and botocore (low-level AWS client library) are available at runtime. Without these packages, any Bedrock operation will fail with an import error.

# With pip
pip install agent-memory-server[aws]

# With uv (used by this project)
uv sync --extra aws

2. Export environment variables¶

# ── AWS credentials ──────────────────────────────────────────────
# (or use an IAM role / AWS CLI profile instead — see "AWS Credentials" below)
export AWS_ACCESS_KEY_ID=your-access-key-id
export AWS_SECRET_ACCESS_KEY=your-secret-access-key
export AWS_REGION_NAME=us-east-1      # Required: used by LiteLLM for Bedrock API calls
export REGION_NAME=us-east-1          # Optional: used by server-side boto3 utilities (model-existence checks)

# ── Bedrock embedding model (bedrock/ prefix REQUIRED) ──────────
export EMBEDDING_MODEL=bedrock/amazon.titan-embed-text-v2:0
export REDISVL_VECTOR_DIMENSIONS=1024   # Must match the embedding model

# ── Bedrock generation models (NO prefix needed) ────────────────
export GENERATION_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0
export FAST_MODEL=anthropic.claude-haiku-4-5-20251001-v1:0

# ── Redis ────────────────────────────────────────────────────────
export REDIS_URL=redis://localhost:6379

3. Start Redis and the server¶

# Start Redis (requires Docker)
docker-compose up redis -d

# Start the memory server
uv run agent-memory api

The REST API is now available at http://localhost:8000 (docs at /docs).

Tip: To also run background tasks (memory extraction, compaction, etc.), start a worker in a second terminal:

uv run agent-memory task-worker

Why Two Region Variables?¶

The server reads the AWS region in two places:

If you only use LiteLLM for Bedrock (the common case), AWS_REGION_NAME is sufficient. Set REGION_NAME as well if you want the server's optional model-existence checks to work. When both are used, set them to the same value.

Understanding the bedrock/ Prefix¶

All LLM operations in the server go through LiteLLM, which uses a provider prefix to route requests to the correct backend.

Why the difference? LiteLLM can infer the provider for generation models from the Bedrock-style model ID (e.g., anthropic.claude-*), so the bedrock/ prefix is optional for generation. Adding it (e.g., bedrock/anthropic.claude-sonnet-4-5-20250929-v1:0) also works. For embedding models, the bedrock/ prefix is the only way LiteLLM distinguishes a Bedrock embedding call from other providers, so it is required. If you omit the prefix on an embedding model, the server will auto-add it and emit a deprecation warning, but this fallback will be removed in a future release.

Supported Models¶

Embedding Models¶

Important: Always use the bedrock/ prefix for embedding models.

* Models marked with * are not in MODEL_CONFIGS and their dimensions cannot be auto-resolved. You must set REDISVL_VECTOR_DIMENSIONS=1024 explicitly when using them, or you will get vector-size mismatch errors at runtime.

Generation Models (Anthropic Claude on Bedrock)¶

The following Anthropic Claude models are available on Bedrock. Models marked pre-configured have entries in MODEL_CONFIGS (in config.py) with validated token limits; the others are fully usable by setting the corresponding environment variable — LiteLLM routes the request based on the Bedrock model ID.

Tip: For the recommended quick-start configuration, use anthropic.claude-sonnet-4-5-20250929-v1:0 as GENERATION_MODEL and anthropic.claude-haiku-4-5-20251001-v1:0 as FAST_MODEL.

Installation¶

AWS Bedrock support requires additional dependencies. Install the [aws] extra:

# With pip
pip install agent-memory-server[aws]

# With uv (recommended — used by this project)
uv sync --extra aws

This installs:

• boto3 — AWS SDK for Python
• botocore — Low-level AWS client library

Without these packages, any attempt to use Bedrock models will fail at import time. The standard install (pip install agent-memory-server) does not include them.

Configuration¶

Environment Variables¶

Configure the following environment variables to use Bedrock models:

# Required: AWS region where Bedrock is available
AWS_REGION_NAME=us-east-1        # Required: for LiteLLM's Bedrock calls
REGION_NAME=us-east-1            # Optional: for server-side boto3 model-existence checks

# For Bedrock Embedding Models (bedrock/ prefix REQUIRED)
EMBEDDING_MODEL=bedrock/amazon.titan-embed-text-v2:0
REDISVL_VECTOR_DIMENSIONS=1024   # Must match the embedding model's output dimensions

# For Bedrock LLM Generation Models (bedrock/ prefix optional)
GENERATION_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0
FAST_MODEL=anthropic.claude-haiku-4-5-20251001-v1:0

# AWS Credentials (choose one method below)

AWS Credentials¶

There are several ways to provide AWS credentials:

Option 1: Environment Variables (Explicit)¶

export AWS_ACCESS_KEY_ID=your-access-key-id
export AWS_SECRET_ACCESS_KEY=your-secret-access-key
export AWS_SESSION_TOKEN=your-session-token  # Optional, for temporary credentials

Option 2: AWS Credentials File¶

The server will automatically use credentials from ~/.aws/credentials:

[default]
aws_access_key_id = your-access-key-id
aws_secret_access_key = your-secret-access-key

Option 3: IAM Role (Recommended for AWS deployments)¶

When running on AWS infrastructure (EC2, ECS, Lambda, etc.), IAM roles provide automatic credential management. LiteLLM will discover credentials from the instance metadata service automatically.

Note: The server's built-in boto3 model-existence utilities (_aws/utils.py) currently require explicit AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables. If you rely solely on an IAM role, LiteLLM Bedrock calls will work, but the optional model-existence checks will be skipped. This limitation may be removed in a future release.

Option 4: AWS SSO / AWS CLI Profile¶

If you've configured AWS SSO or named profiles:

# First, login via SSO
aws sso login --profile your-profile

# The server will use the default profile, or set explicitly
export AWS_PROFILE=your-profile

Docker Configuration¶

The Dockerfile provides two build targets (multi-stage):

• standard (default) — OpenAI / Anthropic support only
• aws — Includes boto3 and botocore for AWS Bedrock support

Building the AWS-enabled Image¶

# Build directly with Docker
docker build --target aws -t agent-memory-server:aws .

Docker Compose¶

The docker-compose.yml ships with a dedicated aws profile that uses pre-built AWS images (redislabs/agent-memory-server-aws). Activate it with --profile aws:

# Start the full AWS stack (API + MCP + task worker + Redis)
docker-compose --profile aws up

# Or start only the API and Redis
docker-compose --profile aws up api-aws redis

Note: The aws profile services (api-aws, mcp-aws, task-worker-aws) are separate from the standard services. Do not mix profiles — run either docker-compose up (standard) or docker-compose --profile aws up.

Create a .env file with your credentials and model configuration. The Docker Compose AWS services read this file automatically:

# AWS credentials
REGION_NAME=us-east-1
AWS_REGION_NAME=us-east-1
AWS_ACCESS_KEY_ID=your-access-key-id
AWS_SECRET_ACCESS_KEY=your-secret-access-key
AWS_SESSION_TOKEN=your-session-token  # Optional, for temporary credentials

# Embedding model (bedrock/ prefix required)
EMBEDDING_MODEL=bedrock/amazon.titan-embed-text-v2:0
REDISVL_VECTOR_DIMENSIONS=1024

# Generation models (override the defaults in docker-compose.yml if desired)
GENERATION_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0
FAST_MODEL=anthropic.claude-haiku-4-5-20251001-v1:0

You can also pass AWS credentials at runtime with docker run:

docker run -e AWS_ACCESS_KEY_ID -e AWS_SECRET_ACCESS_KEY \
  -e REGION_NAME=us-east-1 -e AWS_REGION_NAME=us-east-1 \
  -e EMBEDDING_MODEL=bedrock/amazon.titan-embed-text-v2:0 \
  -e REDISVL_VECTOR_DIMENSIONS=1024 \
  -e GENERATION_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0 \
  -p 8000:8000 redislabs/agent-memory-server-aws:latest

Or mount your AWS credentials directory:

docker run -v ~/.aws:/root/.aws:ro \
  -e AWS_PROFILE=my-profile \
  -e REGION_NAME=us-east-1 -e AWS_REGION_NAME=us-east-1 \
  -p 8000:8000 redislabs/agent-memory-server-aws:latest

Required IAM Permissions¶

The AWS credentials must have permissions to invoke Bedrock models. Here's a minimal IAM policy:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "bedrock:InvokeModel",
                "bedrock:ListFoundationModels"
            ],
            "Resource": "*"
        }
    ]
}

For production, scope down the Resource to specific model ARNs:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": "bedrock:InvokeModel",
            "Resource": [
                "arn:aws:bedrock:us-east-1::foundation-model/amazon.titan-embed-text-v2:0",
                "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-sonnet-4-5-20250929-v1:0",
                "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-haiku-4-5-20251001-v1:0"
            ]
        },
        {
            "Effect": "Allow",
            "Action": "bedrock:ListFoundationModels",
            "Resource": "*"
        }
    ]
}

Note: When using Bedrock LLM models for generation tasks (memory extraction, summarization, topic modeling), ensure your IAM policy includes permissions for all the models you've configured (GENERATION_MODEL, FAST_MODEL, SLOW_MODEL, TOPIC_MODEL).

Vector Dimensions¶

When using Bedrock embedding models, make sure to update the vector dimensions setting to match your chosen model:

# For Titan v2 and Cohere models (1024 dimensions)
REDISVL_VECTOR_DIMENSIONS=1024

# For Titan v1 (1536 dimensions)
REDISVL_VECTOR_DIMENSIONS=1536

Complete Configuration Examples¶

Example 1: Bedrock Embeddings with OpenAI Generation¶

# Embedding model (Bedrock — bedrock/ prefix required)
EMBEDDING_MODEL=bedrock/amazon.titan-embed-text-v2:0
REDISVL_VECTOR_DIMENSIONS=1024

# AWS Configuration
REGION_NAME=us-east-1
AWS_REGION_NAME=us-east-1
AWS_ACCESS_KEY_ID=your-access-key-id
AWS_SECRET_ACCESS_KEY=your-secret-access-key

# Generation model (OpenAI)
GENERATION_MODEL=gpt-4o
OPENAI_API_KEY=your-openai-key

# Other settings
REDIS_URL=redis://localhost:6379

Example 2: Full Bedrock Stack (Recommended for AWS-only deployments)¶

# AWS Configuration
REGION_NAME=us-east-1
AWS_REGION_NAME=us-east-1
AWS_ACCESS_KEY_ID=your-access-key-id
AWS_SECRET_ACCESS_KEY=your-secret-access-key

# Embedding model (Bedrock Titan — bedrock/ prefix required)
EMBEDDING_MODEL=bedrock/amazon.titan-embed-text-v2:0
REDISVL_VECTOR_DIMENSIONS=1024

# Generation models (Bedrock Claude — no prefix needed)
GENERATION_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0
FAST_MODEL=anthropic.claude-haiku-4-5-20251001-v1:0
SLOW_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0

# Other settings
REDIS_URL=redis://localhost:6379

Example 3: OpenAI Embeddings with Bedrock Generation¶

# Embeddings via OpenAI
EMBEDDING_MODEL=text-embedding-3-small
OPENAI_API_KEY=your-openai-key

# Generation via Bedrock
REGION_NAME=us-east-1
AWS_REGION_NAME=us-east-1
AWS_ACCESS_KEY_ID=your-access-key-id
AWS_SECRET_ACCESS_KEY=your-secret-access-key
GENERATION_MODEL=anthropic.claude-sonnet-4-5-20250929-v1:0

REDIS_URL=redis://localhost:6379

YAML Configuration¶

# config.yaml - Full Bedrock Stack
region_name: us-east-1
embedding_model: bedrock/amazon.titan-embed-text-v2:0   # bedrock/ prefix required
redisvl_vector_dimensions: 1024
generation_model: anthropic.claude-sonnet-4-5-20250929-v1:0
fast_model: anthropic.claude-haiku-4-5-20251001-v1:0
slow_model: anthropic.claude-sonnet-4-5-20250929-v1:0
redis_url: redis://localhost:6379

Model Validation¶

The server validates that the specified Bedrock embedding model exists in your configured region at startup. If the model is not found, you'll see an error like:

ValueError: Bedrock embedding model amazon.titan-embed-text-v2:0 not found in region us-east-1.

This helps catch configuration errors early. Common causes:

1. Model not enabled: You may need to enable the model in the Bedrock console
2. Wrong region: The model may not be available in your configured region
3. Typo in model ID: Double-check the model ID spelling

Enabling Bedrock Models in AWS Console¶

Before using a Bedrock model, you must enable it in the AWS Console:

1. Navigate to Amazon Bedrock in the AWS Console
2. Select Model access from the left navigation
3. Click Manage model access
4. Enable the embedding models you want to use
5. Wait for access to be granted (usually immediate for Amazon models)

Mixing Providers¶

You can mix and match providers for embeddings and generation independently. See the Complete Configuration Examples section above for full .env snippets covering each combination.

This flexibility allows you to:

• Keep all data within AWS for compliance requirements
• Use the best model for each task (e.g., OpenAI embeddings + Bedrock generation)
• Optimize costs by choosing appropriate models for different operations

Troubleshooting¶

"AWS-related dependencies might be missing"¶

You need to install the [aws] extra. The standard install does not include boto3:

pip install agent-memory-server[aws]
# or with uv:
uv sync --extra aws

"Missing environment variable 'REGION_NAME'"¶

The server's Settings class reads the region from REGION_NAME. Set it:

export REGION_NAME=us-east-1
export AWS_REGION_NAME=us-east-1   # Also set this for LiteLLM

"Bedrock embedding model not found"¶

1. Verify the model ID is correct
2. Check the model is available in your region
3. Ensure the model is enabled in Bedrock console
4. Verify your IAM permissions include bedrock:ListFoundationModels

"Bedrock LLM model not responding correctly"¶

1. Verify the model ID matches exactly (including version suffix like :0)
2. Check the model is enabled in your Bedrock console
3. Verify your IAM permissions include bedrock:InvokeModel for the specific model
4. Some models may have different regional availability - check AWS documentation

"Error creating chat completion with Bedrock"¶

1. Check that the model ID is correct and the model is enabled
2. Verify your AWS credentials have bedrock:InvokeModel permission
3. Check the request isn't exceeding the model's token limits
4. Review CloudWatch logs for detailed error messages

Credential Errors¶

If you see authentication errors:

1. Verify your AWS credentials are correctly set
2. Check the credentials have the required IAM permissions
3. If using temporary credentials, ensure they haven't expired
4. Try running aws sts get-caller-identity to verify your credentials work

Model-Specific Issues¶

Model IDs: Ensure you're using the Bedrock-specific model IDs (e.g., anthropic.claude-sonnet-4-5-20250929-v1:0) not the direct provider API model IDs.

Embedding Model Prefix: Bedrock embedding models require the bedrock/ prefix (e.g., bedrock/amazon.titan-embed-text-v2:0). Unprefixed names will work but emit a deprecation warning.

LiteLLM Backend: All Bedrock operations use LiteLLM internally, which provides a unified interface for all Bedrock models. Model-specific formatting is handled automatically.

Performance Considerations¶

• Latency: Bedrock API calls may have different latency characteristics than OpenAI
• Rate limits: Check your Bedrock service quotas for your region
• Caching: The server caches model existence checks for 1 hour to reduce API calls
• Cost: Review Bedrock pricing for your chosen embedding model

Related Documentation¶

• LLM Providers - Comprehensive LLM provider guide (recommended)
• Embedding Providers - Embedding model configuration
• Configuration - Full configuration reference
• Custom Memory Vector Databases - Custom memory vector database setup
• Getting Started - Initial setup guide