Why Isn’t My LlamaIndex Tool Streaming Output with streamText() from the AI SDK?

[nikolailehbrink]

I’m using the following setup to integrate @llamaindex/vercel with the AI SDK:

import { openai } from "@ai-sdk/openai";
import { llamaindex } from "@llamaindex/vercel";
import { stepCountIs, streamText } from "ai";
import { LlamaCloudIndex } from "llama-cloud-services";

async function main() {
  const index = new LlamaCloudIndex({
    name: "example",
    projectName: "example",
    apiKey: process.env.LLAMA_CLOUD_API_KEY,
  });
  console.log("Successfully created index");

  const result = streamText({
    model: openai("gpt-4o"),
    prompt: "Tell me about Nikolai's hobbies.",
    toolChoice: "required",
    tools: {
      queryTool: llamaindex({
        model: openai("gpt-4o"),
        index,
        // options: {
        //   fields: ["delta", "stream"],
        // },
        description:
          "Fetch information from your knowledge base to answer questions.",
      }),
    },
  });

  for await (const textPart of result.fullStream) {
    console.log(textPart);
  }
}

main().catch(console.error);

The tool returns the following result:

...,
{
  type: 'tool-result',
  toolCallId: 'call_example123',
  toolName: 'queryTool',
  input: { query: 'Nikolais hobbies' },
  providerExecuted: undefined,
  providerMetadata: {
    openai: { itemId: 'example...' }
  },
  output: "Nikolai's hobbies include... and a lot of other text."
}

The issue:

The tool output isn’t streamed, it only appears once the generation is fully complete.

When I uncomment the options.fields array, the response becomes a structured object instead of plain text, and I noticed one property is stream: false.

So the question is: how can I make the LlamaIndex tool stream its output continuously rather than returning it all at once?