Drop-in AI Transparency & Routing Layer
Mode A is a drop-in Article-12-ready protocol for any LLM API—no model weight access. Mode B adds geometry-gated modules you toggle per workflow.
EU AI Act Article 12 Ready: Designed to support record-keeping and traceability requirements with exportable audit logs and NIST AI RMF transparency practices.
What we mean by "no weight access"
In Mode A we proxy only API calls and never touch model files (pure API operation). In Mode B (self-hosted) we mount customer model files read-only for enhanced telemetry; we never alter model weights. Both modes preserve customer ownership and integrity.
Token-Level Health Monitoring & Routing
Uncertainty, drift, and instability detection with per-request audit trails
We measure token-level health signals (uncertainty, drift, instability proxies) during generation and route traffic to the healthiest provider/expert, creating comprehensive audit logs for compliance.
What We Monitor
- • Token Health: Uncertainty, stability, and quality signals
- • Routing Decisions: Provider selection based on real-time performance
- • Safety Events: Policy triggers, blocks, and security incidents
Compliance Logging
- • Time-stamped request/response hashes and routing decisions
- • Model/provider ID, version, latency, and integrity fingerprints
- • Exportable logs supporting EU AI Act Article 12 requirements
How It Works
Mode A — Observe & Prove
API proxy that captures token-level health signals and emits signed, exportable logs for Article-12/NIST reviews (no model files required).
Mode B — Route & Optimize
Customer-hosted, read-only adapters for deeper telemetry and per-token routing to specialist experts; customer keeps model ownership.
EU AI Act Article 12 Support
Designed to support record-keeping and traceability requirements with exportable event logs over system lifetime and relevant event documentation.
NIST AI RMF Alignment
Transparency and explainability functions aligned with NIST AI Risk Management Framework practices and Generative AI Profile guidance.
Audit Log Schema
Structured JSON logging with timestamp, request ID, model version, routing decisions, safety events, and integrity hashes for compliance.
Retention & Export Controls
Configurable retention policies (6-24 months default) with one-click export to CSV/Parquet for regulators and auditors.
Security Hardening
- • Read-only container FS & read-only weights (when applicable)
- • No model weights required in Mode A (pure API proxy)
- • Least-privilege access; per-request logs only
- • In-VPC / on-prem options; retention 6–24 months configurable
Integration Ready
RESTful APIs with comprehensive documentation for seamless integration into existing AI workflows and SIEM systems.
Ready for EU AI Act compliance?
Get started with transparent AI routing and comprehensive audit logging
Standards Alignment: Aligned to EU AI Act Article 12 record-keeping (provider-controlled, automatically generated logs) and the NIST AI RMF(Govern/Map/Measure/Manage) frameworks.(EUR-Lex),(NIST Publications)
Scope: We provide the governance/observability layer. Article 12 obligations remain with the provider/deployer; our logs and runbooks supply the evidence.