Llama OpenAI Structured Output

At the linked source revision, LlamaOpenAI has a message-formatter signature mismatch and fails before sending a request. Apply the compatible adapter instructions below before running this example.

This current compatible configuration sends a JSON Schema request. Agno then parses the returned content against the Pydantic schema; check its type before reading model fields.

structured_output.py

"""
Meta Structured Output
======================

Cookbook example for `meta/llama_openai/structured_output.py`.
"""

from typing import List

from agno.agent import Agent, RunOutput  # noqa
from agno.models.meta import LlamaOpenAI
from pydantic import BaseModel, Field
from rich.pretty import pprint  # noqa

# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------


class MovieScript(BaseModel):
    setting: str = Field(
        ..., description="Provide a nice setting for a blockbuster movie."
    )
    ending: str = Field(
        ...,
        description="Ending of the movie. If not available, provide a happy ending.",
    )
    genre: str = Field(
        ...,
        description="Genre of the movie. If not available, select action, thriller or romantic comedy.",
    )
    name: str = Field(..., description="Give a name to this movie")
    characters: List[str] = Field(..., description="Name of characters for this movie.")
    storyline: str = Field(
        ..., description="3 sentence storyline for the movie. Make it exciting!"
    )


# Agent that uses a JSON schema output
json_schema_output_agent = Agent(
    model=LlamaOpenAI(id="Llama-4-Maverick-17B-128E-Instruct-FP8", temperature=0.1),
    description="You are a helpful assistant. Summarize the movie script based on the location in a JSON object.",
    output_schema=MovieScript,
)

json_schema_output_agent.print_response("New York")

# ---------------------------------------------------------------------------
# Run Agent
# ---------------------------------------------------------------------------

if __name__ == "__main__":
    pass

output_schema describes the expected type. If parsing or validation fails, result.content can remain a string. Before accessing schema fields in a run result, use isinstance(result.content, YourSchema), replacing YourSchema with the class you passed as output_schema.

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U agno llama-api-client openai

Export your Meta Llama API key

export LLAMA_API_KEY="your_llama_api_key_here"

Use the compatible API adapter

Replace the LlamaOpenAI import (or Llama in the byte-image source) with this helper. Then replace every LlamaOpenAI(...) or Llama(...) construction in the saved file with llama_model(...). Keep the existing id, temperature, and any retry options inside those calls.

Compatible model helper

from os import getenv

from agno.models.openai.like import OpenAILike


def llama_model(**kwargs):
    return OpenAILike(
        api_key=getenv("LLAMA_API_KEY"),
        base_url="https://api.llama.com/compat/v1/",
        supports_native_structured_outputs=False,
        supports_json_schema_outputs=True,
        **kwargs,
    )

This uses Meta's OpenAI-compatible endpoint. You need a Meta API account with access to the selected model; check your account's current model catalog before running.

Run the example

Save the code above as structured_output.py, then run:

python structured_output.py

Full source: cookbook/90_models/meta/llama_openai/structured_output.py