From policy to enforcement in under 120ms
Clauvis is a six-layer engine that transforms raw contracts and policies into real-time, enforceable governance decisions — fast enough to sit inline with a live transaction.
Every agent action routes through one decision point
Intent
A user or workflow asks an agent to act.
Tool call
The agent forms an action — create PO, approve invoice, grant headcount.
Clauvis
The action is intercepted and validated against rules and live state.
Decision
Approve, Block or Escalate — returned with cited evidence.
Execution
Only approved actions reach the system of record.
Follow one purchase order through all six layers
Step through — or press play and watch — a single agent action move from contract ingestion to a sealed, escalated decision.
Worked example — indirect procurement
A procurement copilot raises a EUR 42,000 purchase order against a supplier whose annual cap is EUR 500,000. Follow it through every layer.
The master agreement is parsed
The Meridian Components MSA is ingested. Clauses, spend caps, rebate tiers and notice periods are extracted into structured, queryable terms.
Engine trace
›contract.ingest → MSA-MERIDIAN-2026.pdf (74 pp)
›clauses.extracted = 212
›cap.annual = EUR 500,000 | clause 7.2
Six loosely-coupled layers
Each layer is independently deployable and communicates through defined contracts, so governance scales horizontally alongside your AI adoption.
Legend
Agent node — An AI agent or copilot issuing an action — a PO, invoice release, discount or hire.
Gate node — The evaluation point on the governance band where the action is checked against policy and contract terms.
Approved path — A thread that clears the gate and continues down to commit in the system of record.
Terminated path — A thread stopped at the gate — blocked or escalated, never reaching the system.
Ingestion & Normalisation
Contracts and policy documents are parsed and structured — clauses, entities and tables extracted into a consistent, queryable form.
Contract Intelligence Engine
Clauses are turned into codified, machine-enforceable rules. Ambiguous terms are scored and flagged for human review before they ever govern a transaction.
Digital Contract Twin
A live, stateful model of every obligation and limit — committed spend, remaining balances, capacity and upcoming milestones — so risk is quantified in real time.
Assurance Engine — Core
The decision layer. It validates each agent action against active rules and live twin state, then returns Approve, Block or Escalate with a confidence score.
Immutable Audit Layer
Every decision, its evidence and timestamp are written to a tamper-evident, hash-chained log — ready for regulator export on demand.
Integration & Output
Native hooks into enterprise systems and a human review inbox route decisions to the right place — the transaction, or a named reviewer.
One deterministic decision, every time
For each action, Clauvis returns a single outcome with the evidence, the cited clause and — where relevant — remediation options.
Every policy passes and live risk is within tolerance. The action proceeds to the system of record.
A hard policy or contract term is violated. The action is stopped before execution, with the exact clause cited.
The decision is ambiguous or borderline. It routes to a named human reviewer — a first-class, fully logged decision state.
Put a governance engine between your AI agents and your systems.
See how Clauvis validates agent actions across ERP, CRM, HR, Procurement and Finance — before anything executes.