White paperGoverning AI agents in production: policies, human approvals and an emergency stop.Read the white paper
NavinAI Platform
FRFrançais
Documentation
Request a demo

Governance

Rules enforced by the platform, on every call

Policies, permissions, approvals and registries are held by a single Control Plane. The gateways enforce the decision before every call to a model, a tool or data.
Request a demoContact us

At a glance

  • Stacked policies: platform, tenant, project
  • Four-eyes approvals
  • Kill switch with receipt and controlled restart

The challenge

Without central governance, every project invents its own rules

When every team picks its own models, tools and limits, the organization no longer knows who can access what, what it costs, or how to stop an agent that goes wrong.

  • Different rules from one project to the next
  • Access granted without approval or expiry
  • No fast way to stop an agent, a tool or a provider
  • Costs discovered on the invoice

Policies

One decision, three levels of policy

OPA Rego policies stack from the platform down to the project. The enforcement point built into every gateway evaluates the request and applies the decision.
  1. 01

    Platform

    Operator rules: allowed providers, classifications, ceilings

  2. 02

    Tenant

    Entity rules: models, regions, data, quotas

  3. 03

    Project

    Team rules: tools, MCP, skills, budgets

  4. 04

    Decision

    Allow, deny or request an approval

The decision and its reason are written to the audit trail.

The tools

The building blocks of governance

  • Policy engine

    OPA evaluates versioned Rego policies; an enforcement point in every gateway applies the decision, independently of application code.

  • Fine-grained authorization

    OpenFGA describes who can do what on which resource; Keycloak provides OIDC and SAML authentication, with a dedicated realm or federation per tenant.

  • Approvals

    Sensitive actions go through an approval workflow, with a human in the loop for agents and four-eyes validation for critical operations.

  • Kill switch

    An agent, a tool, a model or a provider stops immediately. Every stop produces a receipt; restarting requires a second validation.

  • Signed execution context

    The Context Authority issues a signed execution context: delegations, mandates and emergency access are bounded and traced.

  • Governance registry

    Models, providers, prices, agents, tools and MCP servers are declared in single registries, with their owner and status.

  • Quotas and budgets

    Quotas and budgets per tenant, project and environment; cost is estimated before the call and measured after.

  • Risk and compliance

    Risks, incidents and their remediation are tracked in the platform and linked to the audit logs.

The proof

Governance you can check in the audit trail

Every policy decision, every approval and every emergency stop leaves a trace your risk, compliance and audit teams can use.

  • The reason behind every policy decision
  • The identity of the approver and of the acting agent
  • Receipts of kill switches and restarts
  • Cost reconciliation per tenant and per project

Go further

  • White paperGovernance

    Governing AI agents in production

    How Navin Enterprise AI keeps autonomous agents under control: OPA policies stacked from platform to tenant to project, approvals with a human in the loop, a kill switch available at any time and a governance registry as the single source of truth.

    25 min read
  • ArticleGovernance

    FinOps for LLM usage: estimated, measured, billed

    Navin Enterprise AI tracks the cost of every model call in three stages: estimated before the call, measured afterwards from the Usage Event, then reconciled with the provider invoice. Token accounting, a pricing registry and quotas keep the cycle under control.

    2 min read

The other capabilities

Put your rules at the heart of every AI call

We build the first version of your policies and approval circuits with you.

Request a demoContact us