Feature Request: Structured Output via Pydantic Models

[sleynsol]

It would be great if the agent framework supported structured output by accepting a pydantic.BaseModel as a schema for the expected result.

---

Motivation

Right now, outputs are unstructured and require manual parsing. With structured output, I could define a schema upfront and let the agent handle formatting + parsing automatically.

---

Example

from pydantic import BaseModel

class AgentResponse(BaseModel):
    success: bool
    output: str | None
    foo: str
    bar: str

agent.set_output_schema(AgentResponse)
response = await agent.run()
assert isinstance(response, AgentResponse)

[rohitdash08]

@sleynsol , I have created the PR #134 for this requirement.
I have tested it locally, please have a look.

[Mohd-Mursaleen]

is it done ? if open i would like to work on it

[RasulOs]

Claude:
This shipped — pass your Pydantic model via output_model and the run result comes back parsed:

from pydantic import BaseModel
from mobilerun import MobileAgent

class AgentResponse(BaseModel):
    success: bool
    output: str | None

agent = MobileAgent(goal="...", output_model=AgentResponse)
result = await agent.run()
result.structured_output  # AgentResponse instance

After the task finishes, a dedicated StructuredOutputAgent (mobilerun/agent/oneflows/structured_output_agent.py) extracts the final answer into your schema. Docs: docs/features/structured-output.mdx. Available since v0.6.0.

Closing as shipped.