afo 0.1.2

Agent Focused Optimization framework for bounded agent optimization loops

AFO

Agent Focused Optimization (AFO) is a small Python framework for bounded agent optimization loops.

AFO separates the agent that chooses actions from the world model that verifies, evaluates, records, and checks convergence. The package is intentionally generic: users provide their own design space, evaluation model, tools, and convergence criteria.

Installation

python -m pip install afo

For local development from this repository:

python -m pip install -e .

Quick Start

Run the bundled quadratic smoke example:

afo run-example quadratic

Or run it from source:

python examples/quadratic/run_quadratic.py

Core Concepts

• Agent: selects the next bounded tool call.
• ActionSpace: declares the tools the agent may use.
• DesignSpace: verifies proposed design variables.
• EvaluationModel: evaluates verified design variables.
• WorldModel: verifies, evaluates, records, and checks convergence.
• RunLoop: executes the default AFO loop.

Minimal Example

from afo.logical.world_model.evaluation_model import EvaluationModel
from afo.logical.world_model.evaluation_result import EvaluationResult


class MyEvaluationModel(EvaluationModel):
    def evaluate(self, design_variable, evaluation_inputs=None):
        return EvaluationResult(value=design_variable["x"] ** 2)

See examples/quadratic/ for a complete runnable example.

SysML

The repository includes AFO.sysml, the SysML source describing the framework requirements, functional actions, logical contracts, and physical flow.

License

AFO is distributed under the GNU General Public License v3.0 only.