Vercel has rolled out several updates to its AI SDK, with changes noted in @ai-sdk/workflow@2.0.14, @ai-sdk/workflow-harness@1.0.93, and @ai-sdk/zai@3.0.2. These updates introduce new capabilities for handling structured output, tool management, and provider-specific enhancements, particularly for Amazon Bedrock.
Key Points
- The AI SDK now exposes parsed structured output in
streamTextend callbacks. - ToolLoopAgent settings and
prepareCallnow permit tool approval secrets. - The AI SDK addresses stateful and empty-match regular expressions in
smoothStream. - Amazon Bedrock integration now accepts citation deltas in streaming responses.
- A model family setting for embeddings has been added to Amazon Bedrock to support ARN.
- Assistant messages are omitted when only cache points remain after unsigned reasoning is filtered in Amazon Bedrock.
Context
According to the Vercel AI SDK Changelog, these updates are part of ongoing development to enhance the SDK's functionality and provider integrations. The changes aim to improve how developers interact with AI models, particularly in streaming scenarios and when utilizing specific provider features like those offered by Amazon Bedrock.
Why It Matters
These updates offer developers more granular control and improved data handling within the Vercel AI SDK. The ability to accept citation deltas in Amazon Bedrock streaming responses and manage tool approval secrets in ToolLoopAgent settings can lead to more robust and transparent AI applications, especially for those requiring detailed attribution or secure tool execution.
What To Do
- Review the Vercel AI SDK documentation for details on
streamTextend callbacks and structured output. - Examine the ToolLoopAgent settings to understand how to implement tool approval secrets.
- Developers using Amazon Bedrock should test the new citation delta acceptance in streaming responses.
- Note the new model family setting for embeddings in Amazon Bedrock for ARN support.
