Sign in
← AI PulseSep 15, 2026

Wire · news · Multi-source brief

Vercel AI SDK Updates Google Integration and Tool Output Serialization

Vercel has released updates to its AI SDK, including fixes for Google Gemini tool calling and serialization of JSON values in tool outputs.

By Illumora Editorial

Source · Sep 15, 2026, 4:12 AM · On Illumora · Sep 15, 2026, 4:24 AM

Media from the primary source — shown here so you can stay on Illumora.

Synthesized from multiple allowlisted primaries on the same event. Lanes →

Brief drafted by Illumora’s editorial model from the linked primary source. Ops desk reviews flagged pieces. How we write →

Read the source →Vercel AI SDK Changelog — Release @ai-sdk/sandbox-just-bash@1.0.111 · vercel/ai
Save

Vercel has released updates to its AI SDK, specifically addressing issues within the @ai-sdk/google package and general tool output serialization. These updates are part of releases such as @ai-sdk/sandbox-just-bash@1.0.111, @ai-sdk/angular@3.0.101, @ai-sdk/code-mode@1.0.58, @ai-sdk/google@4.0.70, @ai-sdk/sandbox-vercel@1.0.111, @ai-sdk/vue@4.0.101, and @ai-sdk/tui@1.0.102.

Key Points

  • The ai@7.0.101 package includes a fix to serialize tool output JSON values.
  • The @ai-sdk/google@4.0.70 package now preserves JSON Schema instead of converting it to OpenAPI schema.
  • A fix in @ai-sdk/google@4.0.70 ensures that Gemini's functionResponse.response field remains an object.
  • Previously, non-object tool results (string, number, array, or null) sent to Gemini could cause a socket closure with code 1007.
  • Non-object tool results are now wrapped under an output key as prescribed by the field's documentation.

Context

According to Vercel, these changes address specific technical challenges encountered when integrating with Gemini models and handling tool outputs. The functionResponse.response field in Gemini expects a google.protobuf.Struct, which is designed to accept only objects. The previous behavior allowed non-object results to be sent unwrapped, leading to connection issues.

Why It Matters

These updates improve the reliability of tool calling when using the Vercel AI SDK with Gemini models. Developers can expect more stable interactions and fewer unexpected disconnections when their tools return non-object JSON values.

What To Do

  • Update to ai@7.0.101 or later to benefit from improved JSON serialization.
  • Update to @ai-sdk/google@4.0.70 or later if using Gemini models with the SDK.
  • Review tool implementations to ensure they align with the updated handling of non-object tool results.
  • Test existing applications with the new SDK versions to confirm compatibility and stability.

Keep Exploring

/atlas/**gemini**-family /techniques/output-schema /techniques/ptcf