Describe the bug
When passing an Output.object({ schema }) to an Agent's generateText method while the Agent also has tools configured, it throws an AI_NoOutputGeneratedError.
It seems that the underlying execution cannot handle tool calling and structured output at the same time, but the error message is vague and it's unclear if this is an intended limitation or a bug in how the model response is parsed.
Steps To Reproduce
Create an Agent with a tool configured.
Call generateText on the Agent, and pass output: Output.object({ schema }) in the options.
Observe the AI_NoOutputGeneratedError error.
Code Snippet:
import { Agent } from '@voltagent/core'
import { Output } from 'ai'
import { z } from 'zod'
const myTool = /* ... valid tool definition ... */
const myAgent = new Agent({
name: 'Test Agent',
instructions: 'Use tools and return a structured response.',
model: /* ... your model (e.g., google('gemini-1.5-flash')) ... */,
tools: [myTool],
})
const schema = z.object({
message: z.string(),
})
// This call throws AI_NoOutputGeneratedError
const result = await myAgent.generateText("Hello, do something and return the schema", {
output: Output.object({ schema }),
})Expected behavior
The Agent should either:
Successfully use the tools, finish its reasoning, and then return the final payload matching the Zod schema.
OR, if combining tools and Output.object is strictly not supported by design, it should throw a clear, descriptive validation error immediately (e.g., "Cannot use Output.object when tools are enabled. Use a two-step workflow instead.") rather than failing parsing with AI_NoOutputGeneratedError.
Packages
@voltagent/core: 2.6.6
ai version 6.0.111
Additional Context
No response
Describe the bug
When passing an Output.object({ schema }) to an Agent's generateText method while the Agent also has tools configured, it throws an AI_NoOutputGeneratedError.
It seems that the underlying execution cannot handle tool calling and structured output at the same time, but the error message is vague and it's unclear if this is an intended limitation or a bug in how the model response is parsed.
Steps To Reproduce
Create an Agent with a tool configured.
Call generateText on the Agent, and pass output: Output.object({ schema }) in the options.
Observe the AI_NoOutputGeneratedError error.
Code Snippet:
Expected behavior
The Agent should either:
Successfully use the tools, finish its reasoning, and then return the final payload matching the Zod schema.
OR, if combining tools and Output.object is strictly not supported by design, it should throw a clear, descriptive validation error immediately (e.g., "Cannot use Output.object when tools are enabled. Use a two-step workflow instead.") rather than failing parsing with AI_NoOutputGeneratedError.
Packages
@voltagent/core: 2.6.6
ai version 6.0.111
Additional Context
No response