Govern every consequential
AI decision. Prove what happened.
Built from Lagos · Independent engineering · Open evidence
Lex is the decision and evidence layer between an AI system and the world. Evaluate model responses and agent actions before they reach users, tools, or production systems — then inspect the receipt.
Explore the Governance ObservatoryC + R + S = 1 · M = min(C,R,S) < τ → Governor fires
Continuity · Reciprocity · Sovereignty — three constitutional pillars
Govern the response. Govern the action.
Lex sits at the boundary where an AI decision becomes visible to a person or executable by a tool.
Before an answer reaches a user.
Evaluate model output for manipulation, drift, and constitutional pressure, then return the answer, an intervention, or a clear refusal.
Try text governance →Before an action reaches a tool.
Intercept file, database, shell, web, and multi-step agent actions before execution, with scope, risk, and trajectory checks.
Explore agent governance →Detect. Correct. Prove.
Lex Aureon turns an invisible safety decision into an inspectable operating path — without retraining the model or hiding the boundary of the evidence.
Detect
Identify constitutional pressure before an output is returned or an agent tool executes.
Correct
Apply the minimum necessary intervention at the control boundary, not after the incident.
Prove
Record the decision, state, and outcome in a SHA-256 audit receipt that can be replayed and verified offline.
The landing page describes the receipt contract without inventing a sample verdict. Real governed receipts are inspectable in the public audit surface.
Latest Runtime Trajectory
This is the current governed state: C + R + S = 1, with M = min(C,R,S) as the canonical margin.
Watch Lex allow, reshape, or block.
The M-score is the governing margin. Choose a scenario and watch C, R, and S move through the simplex as Lex detects pressure, intervenes, and records the transition.
Request received
Balanced state inside the stable interior.
Clear scope · reversible intent
Show transformation
Explain how to improve an accessible landing-page hero without changing the design system.
Response allowed · no intervention required
Strong claims, clearly scoped.
Single-pillar Lyapunov result.
Unrestricted multi-pillar analytical claim stays unproven without a declared drift envelope. Deployed guarantee: the guarded discrete descent invariant.
CBF / Lyapunov certificate for the governed counterfactual: verified in simulation, not an analytical proof.
Go deeper only where you need to.
Connect through the API, MCP, and developer keys.
Inspect tool interception, trajectories, and executed traces.
Read the research boundary, benchmarks, and audit receipts in context.
Explore shared controls, observability, and enterprise access.
Correct before the fact. Prove after.
Try Explorer free, see the control point in action, and move to Sovereign when your AI system reaches production.