Summary
Add support for Amazon Bedrock embedding models as a provider for memorySearch. This would enable users running OpenClaw on AWS (via EC2, ECS, Lambda) to use native AWS embeddings without needing external API keys.
Related Issues
Available Bedrock Embedding Models
| Model ID |
Dimensions |
Max Tokens |
Notes |
amazon.titan-embed-text-v2:0 |
256, 512, 1024 |
8,192 |
Latest, configurable dimensions |
amazon.titan-embed-text-v1 |
1,536 |
8,000 |
Original Titan embeddings |
amazon.titan-embed-image-v1 |
1,024 |
128 |
Multimodal (text + image) |
cohere.embed-english-v3 |
1,024 |
512 |
Via Bedrock marketplace |
cohere.embed-multilingual-v3 |
1,024 |
512 |
Via Bedrock marketplace |
Proposed Config
{
"agents": {
"defaults": {
"memorySearch": {
"enabled": true,
"provider": "amazon-bedrock",
"model": "amazon.titan-embed-text-v2:0",
"dimensions": 1024
}
}
}
}Use Case
The sample-OpenClaw-on-AWS-with-Bedrock deployment already uses Bedrock for LLM inference. Adding Bedrock embedding support would make it a complete AWS-native solution without requiring OpenAI/Voyage API keys.
Authentication
Bedrock embeddings would use the same aws-sdk auth already configured for LLM providers, requiring no additional credentials.
Summary
Add support for Amazon Bedrock embedding models as a provider for
memorySearch. This would enable users running OpenClaw on AWS (via EC2, ECS, Lambda) to use native AWS embeddings without needing external API keys.Related Issues
Available Bedrock Embedding Models
amazon.titan-embed-text-v2:0amazon.titan-embed-text-v1amazon.titan-embed-image-v1cohere.embed-english-v3cohere.embed-multilingual-v3Proposed Config
{ "agents": { "defaults": { "memorySearch": { "enabled": true, "provider": "amazon-bedrock", "model": "amazon.titan-embed-text-v2:0", "dimensions": 1024 } } } }Use Case
The
sample-OpenClaw-on-AWS-with-Bedrockdeployment already uses Bedrock for LLM inference. Adding Bedrock embedding support would make it a complete AWS-native solution without requiring OpenAI/Voyage API keys.Authentication
Bedrock embeddings would use the same
aws-sdkauth already configured for LLM providers, requiring no additional credentials.