Trust Engine Overview
The Trust Engine is the core evaluation layer. It runs all five pillars in parallel for every request.
Architecture
When your application calls POST /api/v1/analyze, the Trust Engine:
Tokenizes and preprocesses the prompt + response pair.
Dispatches all requested pillars in parallel.
Aggregates per-pillar scores using your policy weights.
Determines the enforcement action based on your policy thresholds.
Returns the evaluation result (inline mode) or logs and acts async (background mode).
Latency model
Total evaluation latency equals the slowest single pillar (not the sum). P95 latency target is < 50ms in inline mode with all five pillars running.
Pillar weights
Each pillar contributes to the aggregate Trust Score according to its weight. Default weights are equal (0.2 per pillar). You can adjust weights per policy.