Getting started

CerbIA evaluates content through a configured pipeline:

loaders → preprocessors → SecurityGate (scanners → score aggregator → verdict)

Component	Responsibility
Loaders	Produce entries from configured input sources.
Preprocessors	Transform or derive entries before scanning.
SecurityGate	Runs compatible scanners, applies their actions and severities, and returns a verdict for each entry.
Scanners	Inspect entry content and report a risk score, rationale, and optional match locations.
Score aggregators	Combine the severity-weighted scores of blocking findings for the gate.

Component

Responsibility

Loaders

Produce entries from configured input sources.

Preprocessors

Transform or derive entries before scanning.

SecurityGate

Runs compatible scanners, applies their actions and severities, and returns a verdict for each entry.

Scanners

Inspect entry content and report a risk score, rationale, and optional match locations.

Score aggregators

Combine the severity-weighted scores of blocking findings for the gate.

See Architecture for the complete entry lifecycle and result model.

Install the CLI and validate the repository’s example configuration:

pip install "cerbia[cli]"
cerbia validate examples/cli-usage/config.cerbia.yaml
cerbia scan --config examples/cli-usage/config.cerbia.yaml --text "A short message"

scan needs configured loaders or at least one --text or --file argument. It exits 0 for safe results, 1 for unsafe results, and 2 for configuration or handled execution failures. A missing --file path currently yields no entry, which can result in a safe empty scan.

Continue with Configuration. Packages explains the optional ML, Presidio, and ProtectAI boundaries.