Privacy + data handling
Minimize data, redact logs, and export what auditors need.
How we're different
- Evidence-first workflows (not chat transcripts).
- Policy-gated actions with human approvals.
- Data minimization by design.
Trust rule: AI drafts; humans decide for sensitive outcomes.
Data minimization posture
Store only what is needed
KYBRIQ is designed to store the minimum necessary to run onboarding workflows and prove evidence for audits.
- Evidence links, hashes, and timestamps
- Approval history and rationale
- Deterministic workflow state (tasks, exceptions)
Redaction and logging
Operational logs should avoid raw PII whenever possible.
- PII redaction in logs (roadmap)
- “No citation, no field” enforcement for extracted data
- Exports include evidence references, not invented facts
Model usage boundaries
- LLMs draft: summaries, emails, and structured extraction under strict schema.
- Humans approve sensitive outcomes and any external actions.
- Deterministic policies and checklists gate workflow steps.
Not legal advice. Requirements vary by jurisdiction and institution.
Subprocessors (illustrative)
This list is illustrative for internal planning.
- Object storage (S3-compatible)
- Database (Postgres/SQLite for MVP)
- LLM provider (local model gateway preferred)
Production deployments typically align subprocessors with your existing infrastructure choices.