The Vercel AI SDK has released updates for its @ai-sdk/xai, @ai-sdk/harness-cursor, @ai-sdk/harness-fx, and ai packages. These updates introduce new features and address existing functionalities within the SDK.
Key Points
- The
@ai-sdk/harness-cursor@1.0.0package implements a new Cursor harness adapter. - The
@ai-sdk/harness-fx@1.0.0package implements an fx harness adapter. - The
ai@7.0.79package preserves active text and reasoning parts when another merged UI message stream finishes a step. ai@7.0.79also aligns workflow stream normalization with explicit part end chunks.- The
ai@7.0.79package adds batch completion webhooks, withexperimental_startTextBatchaccepting a webhook. - The
@ai-sdk/xaiprovider now reports image moderation blocks as content policy errors, as noted in the@ai-sdk/xai@3.0.125release. - The
@ai-sdk/workflow@2.0.8package allows approvedgenerateText,streamText, andWorkflowAgentturns to continue with a model-visible tool error when revalidated tool input is invalid.
Context
According to the Vercel AI SDK changelog, these updates include major changes such as the implementation of new harness adapters for Cursor and fx. Patch changes in the ai package focus on stream management and the introduction of batch completion webhooks. The @ai-sdk/xai update specifically addresses how image moderation blocks are reported, classifying them as content policy errors. The @ai-sdk/workflow update, released on August 24, focuses on tool error handling within text generation and workflow agent interactions.
Why It Matters
These updates enhance the Vercel AI SDK's capabilities for developers, offering more robust error handling for content moderation and tool usage, alongside new options for managing AI-generated text streams and batch processing. The introduction of new harness adapters expands the SDK's integration possibilities.
What To Do
- Developers should review the updated documentation for
@ai-sdk/harness-cursorand@ai-sdk/harness-fxto understand the new harness adapters. - Note the changes in
ai@7.0.79regarding workflow stream normalization and the newexperimental_startTextBatchwebhook functionality. - Builders using
@ai-sdk/xaishould be aware of the improved reporting for image moderation blocks as content policy errors. - Test
generateText,streamText, andWorkflowAgentinteractions with invalid revalidated tool input to observe the new error continuation behavior in@ai-sdk/workflow@2.0.8.
Keep Exploring
/techniques/system-user-separation /techniques/output-schema
