Concepts
How usertrust governs every LLM call — the banking patterns, cryptographic proofs, and policy engine under the hood.
usertrust combines financial accounting patterns with cryptographic verification to make every AI agent call auditable and budget-safe. These pages explain the core mechanisms.
Two-phase spend
Banking-style authorization holds for atomic budget enforcement
Audit trail
SHA-256 hash-chained, append-only log for every governed call
Merkle proofs
RFC 6962 inclusion and consistency proofs for public verifiability
Local models
How usertrust classifies local endpoints and meters Ollama, vLLM, and LM Studio at nominal rates
Policy engine
12 field operators, scope globs, and time windows for rule-based governance
Pattern memory
SHA-256 prompt-hash history of model, cost, and outcome; suggestModel() returns the model with the best cost-adjusted success ratio for the caller to route on
Circuit breaker
Per-provider failure isolation with automatic recovery