[BUG]: ToolLoopAgent in AI SDK 6 is not captured in LLM Observability

[makinjc]

Tracer Version(s)

5.81.0

Node.js Version(s)

25.2.15.8

Bug Report

This is the sample code for upcoming V6 SDK (which is to be GA next week): https://v6.ai-sdk.dev/docs/agents/overview

However it is not being captured in LLM Observability. I tried regular stream text, object are captured. This is only a wrapper though and I think it will be a very popular implementation moving forward.

If there is any interim steps I can do here to make them available, I'd be glad to do it as well.

  const weatherAgent = new ToolLoopAgent({
    model: google("gemini-2.5-flash-lite"),
    tools: {
      weather: tool({
        description: 'Get the weather in a location (in Fahrenheit)',
        inputSchema: z.object({
          location: z.string().describe('The location to get the weather for'),
        }),
        execute: async ({ location }) => ({
          location,
          temperature: 72 + Math.floor(Math.random() * 21) - 10,
        }),
      }),
      convertFahrenheitToCelsius: tool({
        description: 'Convert temperature from Fahrenheit to Celsius',
        inputSchema: z.object({
          temperature: z.number().describe('Temperature in Fahrenheit'),
        }),
        execute: async ({ temperature }) => {
          const celsius = Math.round((temperature - 32) * (5 / 9));
          return { celsius };
        },
      }),
    },
    // Agent's default behavior is to stop after a maximum of 20 steps
    // stopWhen: stepCountIs(20),
  });
  
  const result = await weatherAgent.generate({
    prompt: 'What is the weather in San Francisco in celsius?',
  });

Reproduction Code

No response

Error Logs

No response

Tracer Config

No response

Operating System

No response

Bundling

Unsure

[sabrenner]

Hi @makinjc, thank you for opening this issue! I believe this is a feature request at this point, since we're not breaking this aspect of the v6 ai sdk, but just need to support it. I've added an item to get to this as soon as possible, but in the meantime, you should be able to manually wrap it with the LLM Observability SDK

import { llmobs } from 'dd-trace';

// ...

const prompt = 'What is the weather in San Francisco in celsius?';
const result = await llmobs.trace({ kind: 'agent', name: 'myAgent' }, async () => {
  const agentResult = await weatherAgent.generate({
    prompt,
  });

  llmobs.annotate({
    inputData: prompt,
    outputData: agentResult // access a specific property off of this that is the text result
  });

  return agentResult
});

I will keep this issue open and link a feature PR to it once I have a moment to add support for it 😄

[PClmnt]

@sabrenner Thanks so much for the workaround. Just +1'ing for the actual feature and thanks so much for all your hard work!

with the normal streamText calls you get a really nice waterfall. So it would be amazing to have this for the agent!

[Prince-Mendiratta]

@sabrenner Would love to see this!