Python: Support pydantic model as function call result

[Copilot]

Motivation and Context

Fixes #236

Pydantic BaseModel subclasses returned from tool functions were not being serialized correctly. The _prepare_function_call_results_as_dumpable function checked for to_dict() which doesn't exist on Pydantic models.

Description

Added BaseModel isinstance check before the to_dict fallback, calling model_dump() for Pydantic models:

# Now works correctly
class WeatherResult(BaseModel):
    temperature: float
    condition: str

@ai_function
def get_weather(location: str) -> WeatherResult:
    return WeatherResult(temperature=22.5, condition="sunny")

# Returns: {"temperature": 22.5, "condition": "sunny"}

• Modified _prepare_function_call_results_as_dumpable to detect BaseModel subclasses and use model_dump()
• Works recursively for models in lists and dicts
• Added tests for single models, lists, dicts, and nested models

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• api.openai.com
  • Triggering command: /home/REDACTED/work/agent-framework/agent-framework/python/.venv/bin/pytest pytest --cov=agent_framework --cov-report=term-missing:skip-covered tests (dns block)
• astral.sh
  • Triggering command: /usr/bin/curl curl -LsSf REDACTED (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Python: [Python] Support pydantic model as function call result</issue_title>
<issue_description>Currently, Pydantic model as a tool result is not supported.

def _prepare_function_call_results_as_dumpable(content: Contents | Any | list[Contents | Any]) </issue_description>

<agent_instructions>take the _prepare_function_call_results_as_dumpable function and add logic that checks for BaseModel subclasses, if we encounter those, then use content.model_dump() </agent_instructions>

Comments on the Issue (you are @copilot in this section)

@eavanvalkenburg we can do that, I'll add it to my backlog

• Fixes Python: [Python] Support pydantic model as function call result #2511

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[Copilot]

Pull request overview

This PR adds support for Pydantic BaseModel subclasses as function call results in the agent framework's type system. Previously, only objects with a to_dict() method were properly serialized; this change enables Pydantic models to be serialized using their native model_dump() method.

Key Changes:

• Added BaseModel instance check in _prepare_function_call_results_as_dumpable() before the generic to_dict() fallback
• Comprehensive test coverage for single models, lists, dicts, and nested Pydantic models

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/core/agent_framework/_types.py	Added BaseModel check and model_dump() call in the serialization helper function
python/packages/core/tests/core/test_types.py	Added test helper classes and four test functions covering various Pydantic model serialization scenarios

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_types.py	947	99	89%	130–131, 149–150, 287, 289, 296, 315, 355, 401–402, 438, 588, 702–703, 705, 730, 737, 754–756, 829, 834–835, 837, 844–845, 847, 869, 876, 879–881, 886–887, 893–895, 1019, 1106–1109, 1117–1118, 1209, 1390, 1396, 1640–1642, 1648–1649, 1819, 1950, 1955, 1959, 1963, 2144, 2195–2199, 2209, 2214, 2673, 2759–2761, 2834, 2845–2846, 3020, 3024, 3036–3038, 3139–3141, 3143–3145, 3148, 3152, 3155, 3160, 3205–3206, 3213–3214, 3248–3250, 3265, 3281, 3309, 3316
TOTAL	15888	2416	84%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
2286	130 💤	0 ❌	0 🔥	55.944s ⏱️

[eavanvalkenburg]

@copilot there are some code quality checks failing, please fix

[Copilot]

@copilot there are some code quality checks failing, please fix

The Python - Code Quality and Python - Tests workflows are now passing. The CI checks completed successfully after the initial run.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• test.search.windows.net
  • Triggering command: /home/REDACTED/work/agent-framework/agent-framework/python/.venv/bin/pytest pytest --cov=agent_framework_azure_ai_search --cov-report=term-missing:skip-covered tests (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[sebastian-velten]

Hello everyone. I'm happy to see, that pydantic models are now supported as function tool output.
I have two remarks on the changes, made in this issue.

1. Wouldn't it be better to directly return model_dump_json() of the response instead of model_dump() and then dumping it again later using json.dumps(). This method ensures the string output is definitely JSON serializable, which is not the case for the dict resulting from model_dump(). For instance, I think the current implementation crashes if the function call response contains a datetime property.
2. First check fo the "to_dict" method, and then for BaseModel instance. I think, the "to_dict" case is more specific and should therefore have higher priority.

The resulting _prepare_function_call_results_as_dumpable(...) would then be:

def _prepare_function_call_results_as_dumpable(content: Contents | Any | list[Contents | Any]) -> Any:
    if isinstance(content, list):
        # Particularly deal with lists of Content
        return [_prepare_function_call_results_as_dumpable(item) for item in content]
    if isinstance(content, dict):
        return {k: _prepare_function_call_results_as_dumpable(v) for k, v in content.items()}
    if hasattr(content, "to_dict"):
        return content.to_dict(exclude={"raw_representation", "additional_properties"})
    if isinstance(content, BaseModel):
        return content.model_dump_json()
    return content

If you agree with my points (either one of them or both), I would be glad you would assign the issue to me, so I can make the suggested changes and adapt test cases accordingly.