If you build on the Vercel AI SDK, this release adds batch output parsing and extraction model registry support. It also fixes a validation path where deeply nested metadata could exhaust worker memory.
Key takeaways · AI-distilled
Vercel AI SDK 7.0.174 fixes UI message validation that used a recursive schema needing several times the memory of the metadata it checked, so a valid request with deeply nested provider or tool metadata could exhaust a worker's memory before the model was called.
Those metadata values are now validated in a single pass and no longer copied, except to strip __proto__ keys. The change covers validateUIMessages, safeValidateUIMessages, the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → UI stream helpers and citation metadata.
The release adds output support to startBatch, parsed back in getBatchResults, and adds extraction model support to provider registries, custom providers and string model IDs.
streamText no longer passes tools: undefined to experimental_transform factories when no tools are given; transforms now receive an empty tool set, matching the StreamTextTransform type.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.