ai@7.0.135
Vercel AI SDK 7.0.135 introduces a persistent WebSocket chat transport, top-level dimensions settings for embeddings, warnings for tool step limits, and fixes for tool approval requests and stream cleanup.
Why this signal matters
The ai@7.0.135 release introduces several features and bug fixes across core AI workflows. Key additions include a persistent WebSocket chat transport featuring correlated streaming, serialized backpressure, cancellation, validated protocol frames, explicit lifecycle management, and sequence-based resume support. Embeddings now support a top-level 'dimensions' setting, and tool execution loops will log a warning when stopped by the default step limit. Bug fixes prevent duplicate continuation requests on repeated tool approval responses and ensure chat stream cleanup is awaited when stopping. Related workspace dependencies (@ai-sdk/gateway, @ai-sdk/provider, @ai-sdk/provider-utils) were also updated.
Actionable summary
Vercel AI SDK version 7.0.135 adds a persistent WebSocket chat transport supporting sequence-based resume and backpressure, adds a top-level dimensions setting to embeddings, logs warnings on default tool loop limits, and resolves duplicate continuation requests on repeated tool approvals.
- Agent usefulness
- 77/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade to ai@7.0.135 to leverage the new persistent WebSocket chat transport and fix duplicate continuation requests on tool approval.