streamText & useChat: do not stream tool results to UI

[nazarii-klymok]

Hi! I'm using AI sdk 4.0.0 with OpenAI. Here's my code from a back-end. The result is stream to front-end where a useChat hook is used.:

return streamText({
      model: openai(model),
      messages,,
      temperature,
      tools: {
        getSensitiveData: {
          parameters: z.object({ query: z.string() }),
          execute: async ({ query }) => {
            const result = ...
            return result
          },
          description:
            "Tool description"
        }
      },
      maxSteps: 2,
      onFinish: async ({ text }) => {
        ...
      }
    })

From what I can see, the result that I return during tool invocation is also streamed to the UI.
I want the tool result to be generated and used by LLM to generate a response, but I'd prefer to not return tool response back to the client. Is it possible with streamText or any other AI API?

Thank you.

[nicoalbanese]

Hey! Rather than having that logic as a tool, have you thought about having it inline in whatever tool needs the sensitive data?