What should we do?
Replace the LangChain4j bridge for the bedrock provider discriminator with a native implementation over the AWS SDK v2 bedrockruntime Converse API. Supports both Anthropic and non-Anthropic models on Bedrock (Nova, Mistral, Llama, Cohere, Claude, …) via the generic Converse adapter — matching today's LangChain4j routing. The lossless Anthropic-on-Bedrock path via the Anthropic SDK platform-backend module is a separate Phase 2 issue, available only via the new connectors.
Requirements:
- Streaming-first internal driver.
- Hardcoded
ModelCapabilities: text + image + document on user-message (where supported by the family); text on tool-result; text on assistant-message.
- AWS auth chain (existing credential resolution) and AWS SDK HTTP client with Apache transport + proxy via the shared util.
- Stop-reason mapping, token usage, error classification consistent with the other native impls.
Why should we do it?
Day-one feature parity with the LangChain4j implementation for AWS Bedrock customers.
What should we do?
Replace the LangChain4j bridge for the
bedrockprovider discriminator with a native implementation over the AWS SDK v2bedrockruntimeConverse API. Supports both Anthropic and non-Anthropic models on Bedrock (Nova, Mistral, Llama, Cohere, Claude, …) via the generic Converse adapter — matching today's LangChain4j routing. The lossless Anthropic-on-Bedrock path via the Anthropic SDK platform-backend module is a separate Phase 2 issue, available only via the new connectors.Requirements:
ModelCapabilities: text + image + document on user-message (where supported by the family); text on tool-result; text on assistant-message.Why should we do it?
Day-one feature parity with the LangChain4j implementation for AWS Bedrock customers.