Bedrock embeddings now supported for memory search
Memory search can now use Amazon Bedrock embedding models (Titan, Cohere, Nova, TwelveLabs) with automatic AWS credential detection. No API key configuration needed on EC2, ECS, or Lambda deployments.
When OpenClaw runs on AWS infrastructure, developers previously had to configure external embedding APIs like OpenAI or Voyage AI — even when Bedrock was already handling inference. A new embedding provider bridges that gap.
Memory search now supports Amazon Bedrock as a native embedding backend. The provider auto-detects when AWS credentials are available via the standard credential chain (instance roles, environment variables, SSO, shared config) and activates automatically in provider: "auto" mode. No API keys are required — the same IAM role that grants Bedrock inference access handles embeddings too.
Supported models include Amazon Titan Embed Text (v1 and v2), Cohere Embed (v3 and v4), Amazon Nova 2 Multimodal Embeddings, and TwelveLabs Marengo. Titan V2 supports configurable output dimensions (256, 512, or 1024) through the existing outputDimensionality setting.
The implementation lives in the memory-host-sdk package, using the AWS SDK's BedrockRuntimeClient for API calls and the credential-provider-node package for credential resolution. Model families have different request and response shapes, so the provider handles family-specific JSON formatting and parsing accordingly.
View Original GitHub Description
Add Amazon Bedrock as a native embedding provider for memory search. Supports Titan Embed Text v1/v2 and Cohere Embed models via AWS SDK.
- New embeddings-bedrock.ts: BedrockRuntimeClient + InvokeModel
- Auth via AWS default credential chain (same as Bedrock inference)
- Auto-selected in 'auto' mode when AWS credentials are detected
- Titan V2: configurable dimensions (256/512/1024), normalization
- Cohere: native batch support with search_query/search_document types
- 16 new tests covering all model types, auth detection, edge cases
Closes #26289