Compliance tool

Introduced 0.13.0

The ComplianceTool, which an agent sees as CheckCompliance, reports what is sensitive in a piece of text and what a named compliance profile applies to those matches.

The answer lists kinds, counts, and character offsets. The matched substrings are omitted from the answer, the transcript, the log, and the model's context window.

Detectors match patterns. A result with no matches means those detectors found none in this text. Use it before quoting, summarising, or forwarding content you are unsure about. Release decisions for imagery belong to CheckDissemination.

Parameters

Parameter	Required	Description
content	Yes	The text to examine.
profile	No	The regime to apply: pii_basic (default), gdpr, ccpa, hipaa, pci_dss, soc2, or fedramp. See the compliance processor for the full catalog.
redact	No	When true, the answer also includes the redacted text.

Audit trail

Every decision is written to .lucenia-compliance-audit, including passes. Verdicts are refused, no_refusal, and undetermined. The record stores a digest of the content and the shape of what was found. If the write fails, the tool answer says the decision stands and the record does not exist.

Register a flow agent that runs the ComplianceTool

POST /_plugins/_agent/agents/_register
{
  "name": "Compliance_Agent",
  "type": "flow",
  "description": "checks text for sensitive data",
  "tools": [
    {
      "type": "ComplianceTool",
      "name": "DemoCheckCompliance",
      "parameters": {
        "profile": "hipaa"
      }
    }
  ]
}

POST /_plugins/_ml/agents/<agent_id>/_execute
{
  "parameters": { "content": "the text to examine" }
}