Llama Cpp Structured Output

Request a MovieScript with JSON Schema and validate the model response locally.

structured_output.py
"""
Llama Cpp Structured Output
===========================

Cookbook example for `llama_cpp/structured_output.py`.
"""

from typing import List

from agno.agent import Agent
from agno.models.llama_cpp import LlamaCpp
from agno.run.agent import RunOutput
from pydantic import BaseModel, Field
from rich.pretty import pprint  # noqa

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


class MovieScript(BaseModel):
    name: str = Field(..., description="Give a name to this movie")
    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.",
    )
    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 returns a structured output
structured_output_agent = Agent(
    model=LlamaCpp(id="ggml-org/gpt-oss-20b-GGUF"),
    description="You write movie scripts.",
    output_schema=MovieScript,
)

# Run the agent synchronously
structured_output_response: RunOutput = structured_output_agent.run("New York")
pprint(structured_output_response.content)

# ---------------------------------------------------------------------------
# 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 openai

Install llama.cpp

Install the llama-server binary. This command supports macOS and Linux with Homebrew; see the llama.cpp installation guide for other platforms:

brew install llama.cpp

Start llama.cpp

Serve ggml-org/gpt-oss-20b-GGUF at http://127.0.0.1:8080/v1:

llama-server -hf ggml-org/gpt-oss-20b-GGUF --ctx-size 0 --jinja -ub 2048 -b 2048

Run the example

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

python structured_output.py

Full source: cookbook/90_models/llama_cpp/structured_output.py