Safe Context & Uncertainty

When using LLMs, it's common to encounter situations where the model produces uncertain or ambiguous responses. OpenHosta provides a "safe context" to handle ambiguity without breaking your code or returning unhandled exceptions deep in the stack.

safe Context Manager

You can define specific conditions under which the model should respond securely or limit the cumulated_uncertainty threshold.

from enum import StrEnum
from OpenHosta import emulate, safe, UncertaintyError

class NextStep(StrEnum):
    GIT_PUSH = "git push"
    GIT_COMMIT = "git commit"
    GIT_STATUS = "git status"

def get_next_step(command: str) -> NextStep:
    """Determines the next logical git step based on the provided command."""
    return emulate()

if __name__ == "__main__":
    with safe(acceptable_cumulated_uncertainty=1e-5) as safety_context:
        try:
            for i in range(3):
                next_step = get_next_step("git commit -m 'Initial commit'")

            # Trigger high uncertainty 
            next_step = get_next_step("git is forbidden here")

        except UncertaintyError as e:
            print(f"Caught expected UncertaintyError: {e}")

Handling errors securely guarantees that unexpected hallucinated output won't propagate logic bugs in production scripts.