Every LLM hallucinates. TriniGard sits between your data and your model, verifying claims against authoritative sources before they reach your users.
TriniGard is not a prompt guardrail. It is a governance layer that verifies factual claims against your own data.
Query PostgreSQL, REST APIs, and Elasticsearch in parallel. Cross-reference claims against your authoritative data sources, not just pattern matching.
Per-use-case confidence levels from 0.65 (customer support) to 0.98 (aviation safety). Set your risk tolerance, not ours.
Write-ahead logging with HMAC signing and async DB sync. Every verification decision is cryptographically logged for SOC2/HIPAA compliance.
Abstract adapter pattern supports any data source or LLM provider. Switch from Claude to GPT to your own model without changing verification logic.
Redis-cached verification results, circuit breakers, and parallel source queries. Sub-300ms verification in typical deployments.
React-based web UI for use case management, live verification testing, audit review, and system health monitoring.
Three-tier roles (admin/editor/viewer) with JWT authentication. Control who creates use cases, runs verifications, and manages system settings.
Isolate teams, departments, or customers with per-tenant configuration, quotas, and independent trust policies.
Generate compliance reports covering 10 control checks including access control, audit logging, data retention, and change management.
Configure automatic cleanup of audit logs and verification results per tenant. Stay compliant with GDPR, HIPAA, and internal data policies.
Integrated Prometheus metrics, Grafana dashboards, and Alertmanager alerts. Track verification latency, error rates, and system health in real time.
GitHub Actions for automated testing, linting, security scanning, and Docker image builds on every push and pull request.
Each industry has different risk tolerances. TriniGard adapts to yours.
| Vertical | Confidence Threshold | Max Discrepancies | Required Sources | Fallback |
|---|---|---|---|---|
| Healthcare (Medical Diagnosis) | 0.97 | 1 | 3 | reject |
| Aviation Safety | 0.98 | 0 | 3 | reject |
| Financial Compliance | 0.95 | 1 | 2 | flag_for_review |
| Legal Contract Review | 0.92 | 1 | 2 | flag_for_review |
| Gaming Regulatory | 0.93 | 1 | 2 | flag_for_review |
| Customer Support (Low Risk) | 0.65–0.80 | 2–5 | 1 | proceed_with_warning |
TriniGard inserts itself between your LLM and your users. Simple architecture, powerful verification.
Send LLM-generated text to TriniGard along with the use case ID.
TriniGard queries configured data sources in parallel: PostgreSQL, APIs, Elasticsearch.
Confidence score computed against your threshold. Pass, flag, or reject.
Every decision logged to WAL with HMAC signature, synced to PostgreSQL.
Try the live demo or read the executive one-pager.