Python: [Bug]: agent-framework-foundry: Cannot use response_format at runtime (run/options) - raises 400 error

[mhmtkcm]

Description

Summary

When using agent-framework-foundry, providing response_format at runtime (e.g., via run() options) results in an error. With the previous approach (AzureAIProjectAgentProvider.get_agent() / create_agent()), we could pass response_format through the options parameter and receive an already-parsed response.

---

Background

Previously, via AzureAIProjectAgentProvider, we could:

• Create an Agent if it didn’t exist (create_agent()), or reuse an existing one (get_agent()).
• Call run() and specify response_format in runtime options, and the returned response would be parsed for us.

After moving to agent-framework-foundry, Agents are only used (creation is delegated to azure-ai-projects). However, we can no longer achieve the same runtime response_format behavior.

---

Problem

If response_format is provided at runtime, the call fails with an API/validation error (the same underlying limitation discussed in previous issues). We understand this may be an API limit, but the parsing does not need to happen server-side. The agent response could simply return the raw content in the text field, and the SDK/client could handle parsing locally.

---

Related Issues

• Python: [Bug]: AzureAIProjectAgentProvider Foundry agents ignore / don’t accept response_format at runtime (agent.run) #3230
• Python: [Bug]: If response doesn't match the JSON schema of AzureAIProjectAgentProvider Foundry V2 agent, response.value returns None #3268

---

Why This Matters

Our integration relies on specifying output structure at runtime (for different tasks/validators). Without runtime response_format, we must implement custom parsing/validation outside the SDK, losing the convenience and consistency we had before.

Code Sample

import logging

from agent_framework_foundry import FoundryAgent
from azure.ai.projects.aio import AIProjectClient
from azure.ai.projects.models import (
    PromptAgentDefinition,
    PromptAgentDefinitionTextOptions,
    TextResponseFormatJsonSchema,
)
from azure.core.exceptions import ResourceNotFoundError
from azure.identity.aio import (
    AzureCliCredential,
    ChainedTokenCredential,
    ManagedIdentityCredential,
)
from pydantic import BaseModel

logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)

for noisy in (
    "agent_framework",
    "azure",
):
    logging.getLogger(noisy).setLevel(logging.WARNING)

AZURE_AI_PROJECT_ENDPOINT = "..."
AGENT_NAME = "TEST-AGENT"
MODEL = "gpt-5.4"
INSTRUCTION = "You are a helpful assistant for accounting questions."


class AgentResponseModel(BaseModel):
    answer: str


async def main():
    async with (
        ChainedTokenCredential(
            ManagedIdentityCredential(),
            AzureCliCredential(),
        ) as credential,
        AIProjectClient(
            endpoint=AZURE_AI_PROJECT_ENDPOINT,
            credential=credential,
            allow_preview=True,
        ) as client,
    ):
        try:
            agent_project = await client.agents.get(agent_name=AGENT_NAME)
            agent = FoundryAgent(
                project_client=client,
                agent_name=AGENT_NAME,
                agent_version=agent_project.versions.latest.version,
            )
            logger.info("Existing agent found: %s", agent.client.agent_name)
        except ResourceNotFoundError:
            agent_project = await client.agents.create_version(
                agent_name=AGENT_NAME,
                definition=PromptAgentDefinition(
                    model=MODEL,
                    instructions=INSTRUCTION,
                    text=PromptAgentDefinitionTextOptions(
                        format=TextResponseFormatJsonSchema(
                            name=AgentResponseModel.__name__,
                            schema=AgentResponseModel.model_json_schema(),
                            strict=True,
                        )
                    ),
                ),
            )
            agent = FoundryAgent(
                project_client=client,
                agent_name=AGENT_NAME,
                agent_version=agent_project.version,
            )
            logger.info("New agent created: %s", agent.client.agent_name)

        response = await agent.run(
            "What is the tax rate for small businesses?",
            options={"response_format": AgentResponseModel},
        )
        logger.info("Agent response: %s", response.value)


if __name__ == "__main__":
    import asyncio

    asyncio.run(main())

Error Messages / Stack Traces

agent_framework.exceptions.ChatClientException: <class 'agent_framework_foundry._agent._FoundryAgentChatClient'> service failed to complete the prompt: Error code: 400 - {'error': {'code': 'invalid_payload', 'message': 'Not allowed when agent is specified. [Request ID: ...]', 'param': 'text', 'type': 'invalid_request_error', 'details': [], 'additionalInfo': {'request_id': '...'}}}

Package Versions

agent-framework-core: 1.1.1, agent-framework-foundry: 1.1.1, azure-ai-projects: 2.1.0

Python Version

Python 3.13.13

Additional Context

No response