-
Notifications
You must be signed in to change notification settings - Fork 1.2k
Open
Labels
enhancementNew feature or requestNew feature or request
Description
🚀 Describe the new functionality needed
The current AWS Bedrock provider uses direct boto3 API calls which creates complexity around credential management and doesn't leverage the existing OpenAI-compatible infrastructure. We need to refactor the Bedrock provider to use Bedrock's OpenAI-compatible endpoint instead.
- Refactor BedrockInferenceAdapter to inherit from LiteLLMOpenAIMixin (like the Groq provider)
- Switch from boto3 dependency to litellm
- Update configuration to use simple API key + region instead of AWS credentials
- Use Bedrock's OpenAI endpoint: https://bedrock-runtime.{region}.amazonaws.com/openai/v1
- Add unit tests for the new implementation
- Update integration tests to remove Bedrock from OpenAI compatibility exclusions
💡 Why is this needed? What if we don't build it?
- Simplified credential management: Users can use API tokens instead of managing AWS access keys
- Consistency: Follows the same pattern as other OpenAI-compatible providers (Groq, etc.)
- Reduced complexity: Eliminates custom boto3 integration code in favor of proven OpenAI compatibility layer
If we don't build it:
- Bedrock remains inconsistent with other providers' authentication patterns
- More complex credential setup for users
Other thoughts
-
Follows established patterns already proven with the Groq provider
-
Leverages existing LiteLLMOpenAIMixin infrastructure
-
Breaking change: Existing configurations will need to be updated from AWS credentials to API key format
-
Dependency change: Switches from boto3 to litellm (but litellm is already used by other providers)
Metadata
Metadata
Assignees
Labels
enhancementNew feature or requestNew feature or request