Feature Guide
Customize how PillarShield governance assesses your content. Control PII detection, prohibited terms lists, and advanced LLM-based analysis from your dashboard.
Governance
Governance settings control the baseline content safety rules.
PII Detection
Enable precise, deterministic scrubbing for common sensitive data patterns.
- Coverage: Email addresses, phone numbers, SSNs, credit card numbers.
- Effect: Flags content as a violation if these patterns are found.
Prohibited Terms
Block specific words or phrases from being published.
- Core Plan: Uses the curated PillarShield baseline list (not editable).
- Protect Plan: You own this list. Add custom terms (one per line) to block competitors, embargoed projects, or internal jargon.
LLM Assistance
Available on all plans, with enforcement checks on Protect.
Tone Detection
Uses a Large Language Model to analyze the sentiment and tone of content.
Directives & Goals
You can set up to 5 simple tone directives, such as:
"Be professional and calm""Avoid sales-heavy language"
Enforcement Thresholds Protect
By default, LLM checks are informational. On the Protect plan, you can turn them into blocking gates by setting thresholds:
| Setting | Description | Recommendation |
|---|---|---|
| Min Score | 0–100. Content scoring below this compliance level triggers a violation. | Start at 70–80. |
| Min Confidence | 0.0–1.0. Analysis is ignored if the model is not at least this confident. | Keep high (0.8+) to avoid false positives. |
PII Redaction Protect
For high-security environments, you can ensure that sensitive data never leaves the API response payload.
- Redact Content: Replaces detected PII in the API response (e.g.,
user@example.com→[EMAIL]). - Safe List: Define specific emails or phone numbers (one per line) that should be ignored and allowed to pass through checks.