Route decisions to the cheapest sufficient layer
Not every query needs a multi-agent debate. Trust Cascade automatically routes each request to the most cost-effective processing tier that can handle it reliably.
Rules Engine
70%Deterministic logic for known patterns. FAQ lookups, policy checks, simple routing.
$0.0001/queryML Models
20%Classification, scoring, pattern matching. Fraud signals, sentiment, categorization.
$0.001/querySingle Agent
8%Reasoning required but straightforward. Research, analysis, content generation.
$0.05/queryMulti-Agent
2%High stakes, verification needed. ARTEMIS debates, consensus protocols.
$0.50/queryThe Math That Matters
A pure multi-agent approach costs ~$0.50/query. Trust Cascade's blended cost: ~$0.07/query. At 1M queries/month, that's $430K saved annually—while maintaining 96% accuracy on the queries that matter most.
How queries get classified
Confidence Scoring
Each tier returns a confidence score. If confidence is below threshold, the query escalates to the next tier automatically.
Stakes Classification
Business rules define what's high-stakes: transaction amounts, customer segments, regulatory implications. High stakes skip to appropriate tiers.
Pattern Learning
The router learns from outcomes. Queries that repeatedly escalate get routed to higher tiers earlier. Efficiency improves over time.
Override Policies
Compliance and risk teams define mandatory escalation rules. Certain query types always go to human review regardless of confidence.