Vercel has updated its AI SDK, with releases for @ai-sdk/workflow-harness@1.0.93 and @ai-sdk/zai@3.0.2. These updates introduce new functionalities and fixes across various components of the SDK, including specific enhancements for Amazon Bedrock integration and general improvements to tool handling and text streaming.
Key Points
- The AI SDK now exposes parsed structured output within
streamTextend callbacks. - ToolLoopAgent settings and
prepareCallnow support tool approval secrets. - The SDK addresses stateful and empty-match regular expressions in
smoothStream. - Amazon Bedrock integration now supports a model family setting for embeddings to accommodate ARN.
- Amazon Bedrock streaming responses can now accept citation deltas.
- Assistant messages are omitted when only cache points remain after unsigned reasoning is filtered within Amazon Bedrock.
Context
According to Vercel, these updates are part of ongoing development, with packages automatically published to npm upon merging changes. The updates affect several core components, including @ai-sdk/gateway@4.0.68, @ai-sdk/provider-utils@5.0.33, @ai-sdk/alibaba@2.0.38, and @ai-sdk/amazon-bedrock@5.0.67.
Why It Matters
These updates enhance the capabilities and reliability of the Vercel AI SDK, offering developers more robust tools for integrating AI models, particularly with Amazon Bedrock. The improvements to ToolLoopAgent and streamText callbacks can streamline development workflows and improve the handling of complex AI interactions.
What To Do
- Review the updated documentation for
streamTextend callbacks to utilize parsed structured output. - Check ToolLoopAgent settings for new options related to tool approval secrets.
- Note the changes in Amazon Bedrock integration, especially the new model family setting for embeddings.
- Observe how Amazon Bedrock streaming responses now handle citation deltas.