cerbia-core | CerbIA

cerbia-core provides cerbia.core, the configurable pipeline used by CerbIA integrations. It includes entries, loaders, preprocessors, scanners, score aggregators, configuration models, Runner, and SecurityGate.

Install

pip install cerbia-core

The core package requires Python 3.12 or later and Pydantic. It does not require the optional ML or NLP runtimes; those capabilities are provided by separate distributions.

Create and run a pipeline

This example creates a text loader, a keyword scanner, and a maximum score aggregator, then scans the configured entry:

from cerbia.core.config import (
    CerbIAConfig,
    LoaderConfig,
    ScannerConfig,
    ScoreAggregatorConfig,
)
from cerbia.core.runner import Runner

config = CerbIAConfig(
    name="inline",
    loaders=[LoaderConfig(loader="cerbia.core.loaders.TextLoader", init_args={"texts": "hello"})],
    scanners=[ScannerConfig(scanner="cerbia.core.scanners.KeywordScanner")],
    score_aggregator=ScoreAggregatorConfig(
        score_aggregator="cerbia.core.score_aggregators.MaxScoreAggregator"
    ),
)
result = Runner(config).scan()

Runner requires at least one loader and scanner. It applies configured preprocessors in order, then evaluates each resulting entry through the gate. result.entry_results contains per-entry results, and result.is_safe reports whether the complete run is safe.

Find component details

Use the product-wide references for the pipeline lifecycle, configuration format, loaders, preprocessors, scanners, and score aggregators. The quickstart walks through a working CLI example.