Re: the AI SDK’s streaming stuff.

I’m using mergeIntoDataStream to pipe results to my client (similar to the example here https://mastra.ai/en/docs/frameworks/agentic-uis/ai-sdk#createdatastream), but I noticed it always includes tool-call and tool-result chunks in the stream. There’s no way to turn those off (unlike the flags for reasoning or sources).

For my use case, I just wanna send the final text and maybe usage, but I don’t wanna leak all the tool-calling internals to the frontend.

Right now, I’m working around it by basically re-implementing the merge logic (stole the code from https://github.com/vercel/ai/blob/main/packages/ai/core/generate-text/stream-text.ts#L1629) and filtering out those chunk types myself, but it feels hacky.

Is there a reason those are always included?

Would it be aligned with your pipeline goal if we introduce chunk filtering mechanism or overriding formatDataStreamPart in it? Basically I am trying to give more freedom to opt-in or out of this different chunktype.

Happy to PR this if it’s something you’d want!
Any suggestions is appreciated, Thanks a ton! 🙏