Live System Metrics
332
API Endpoints
31K
Lines of Code
2
DOI Publications
10
ATF Invariants
0
Violations in Production
Runtime Execution Flow
AI Agent
requests action
→
VeriSigil Gate
intercepts
→
5 Checks
trust · policy · risk
→
ALLOW
action executes
/
DENY
blocked + sealed
/
HUMAN
supervisor required
→
SAC Record
sealed · verifiable
Governance Stack — 8 Layers
Layer 8 — Operating System
Governance OS Kernel
Unified orchestration runtime — policy engine, calibration, sovereign isolation, governance memory
14 endpoints
Layer 7 — Security
AI Security Layer
Adversarial simulator, risk-based governance engine, fail-safe DENY, attack pattern library
5 endpoints
Layer 6 — Threat Defense
Advanced Threat Defense
Inference corruption, semantic integrity, cognition poisoning, provider collapse, model registry
12 endpoints
Layer 5 — Interoperability
ATF ↔ VGS Bridge
Cross-protocol governance — ATF field spec v1.4, SPV binding, TAR↔TAP, RCR↔SAC, all invariants
5 endpoints
Layer 4 — Accountability
Sovereign Accountability Chain VGS-024
5 accountability layers — continuity, supervisor visibility, consequence binding, attribution graph, independent verifier
9 endpoints
Layer 3 — Intelligence
Agent Intelligence — VGS-020 to VGS-023
Agent inventory, shadow detection, topology mapping, governance dashboard, trust trajectory
21 endpoints
Layer 2 — Integrity
Document & Inference Integrity
Semantic drift, clause mutation, intent corruption, numerical inconsistency, inference corruption detection
8 endpoints
Layer 1 — Core
Runtime Enforcement Core — VGS-000 to VGS-019
Runtime gate, EAT tokens, trust scoring, audit chain, jurisdiction, escalation, fail-safe DENY by default
229 endpoints
Legal Invariants — VGS-024
VER-INV-024
Accountability Continuity
Authority was valid and unbroken throughout the entire execution lifecycle — not just at initiation. Critical for legal disputes around revocation timing.
VER-INV-025
Non-repudiation of Authority
Authority is cryptographically bound to the agent, action, and timestamp. The authorizing party cannot deny having approved the action.
VER-INV-026
Supervisory Visibility Fidelity
Reconstructs exactly what the supervisor knew at the moment of authorization. Proves informed consent — not just that approval occurred.
VER-INV-027
Irreversible Consequence Binding
Binds what real-world effect became irreversible. Not what happened — what cannot be undone. Critical for finance, healthcare, defense.
Stakeholder Language Matrix
| Stakeholder | Technical Concept | What It Means To Them |
|---|---|---|
| Board of Directors | Runtime enforcement gate | Every AI decision is reviewed and approved before it takes effect — automatically, at scale |
| CISO | Fail-safe DENY posture | If our governance system goes offline, AI agents automatically stop — fail-closed, not fail-open |
| Chief Risk Officer | Consequence binding VER-INV-027 | We know exactly what became irreversible and when — before litigation, not during |
| Regulator | Independent verifier package | Records can be verified without trusting VeriSigil — no platform dependency for compliance evidence |
| Government | Sovereign isolation mode | AI agents operate within sovereign boundaries — no external telemetry, local trust computation |
| Enterprise Architect | 332 API endpoints + SDK | Drop into any Python stack in 15 minutes — LangChain, CrewAI, AutoGen, LangGraph native |
Agent Framework Connectors
LangChain
Subclass VeriSigilLangChainTool or use @governed_langchain_tool decorator. Every tool call governed before execution.
VeriSigilLangChainTool
CrewAI
VeriSigilCrewAIAgent wraps any CrewAI agent. Governance runs before every task execution. Denied tasks blocked automatically.
VeriSigilCrewAIAgent
AutoGen
Govern functions, messages, and conversation initiation. Works with AssistantAgent, UserProxyAgent, and GroupChat.
VeriSigilAutoGenAgent
LangGraph
Govern StateGraph nodes and conditional edges. On DENY, injects governance_denied into state rather than raising — clean graph flow.
VeriSigilLangGraphNode
Universal
@govern() decorator works on any Python function — framework agnostic. One import, 15 minutes, any stack.
@govern(api_key=..., consequence=...)
5 Threat Defense Coverage
Model Wars
Model governance registry detects unauthorized model switching and regulatory arbitrage. ATF↔VGS bridge enables cross-framework interoperability.
POST /v1/models/register
Provider Collapse
Trust trajectory monitoring, Governance Survivability Index, collapse risk scoring, automatic agent suspension on provider degradation.
GET /v1/governance/collapse-risk
Inference Corruption
Entropy anomaly detection, PII leakage, semantic drift from input, memorization risk — caught before action executes.
POST /v1/inference/verify
Semantic Fragmentation
Context poisoning detection, meaning drift across agent chains, authority reinterpretation detection, intent preservation scoring.
POST /v1/semantic/chain-verify
Cognition Poisoning
Synthetic data signature detection, training data memorization risk, behavioral drift from baseline, entropy collapse monitoring.
POST /v1/cognition/output-integrity
Formal Publications
VeriSigil Governance Stack — Formal Specification
TLA+ verified formal specification of the VeriSigil Governance Stack architecture, invariants, and operational semantics.
Sovereign Accountability Chain — Technical Paper
Runtime enforcement architecture for autonomous AI governance. Includes ATF bridge validation results — 4/4 traces, zero violations.