AI Systems
Data Governance for AI

A policy decision point for the AI you already run

You wrote the routing policy: which data classes may go to which models, under which conditions. The Model Clearance API makes that policy answerable at runtime — your LiteLLM gateway, OpenClaw agent or n8n workflow asks first, gets the deterministic verdict with reasons and alternatives, and every answer becomes evidence. The division of labour is deliberate: Fronterio decides and documents; your infrastructure enforces.

3 verdicts

Green, yellow or red — deterministic, no AI in the decision loop

Append-only

Every decision logged as evidence, visible on the vendors screen

0 payloads

Metadata only — Fronterio never routes, proxies or reads your traffic

From written policy to answerable policy

Most routing policies are documents: correct on the day they were signed, invisible to the systems making the actual calls. The Model Clearance API turns yours into infrastructure — a policy decision point (PDP) your runtimes query before using a model.

The verdict is the same deterministic evaluation as the dashboard's traffic-light preview: agent rules override org defaults, deny beats conditional beats allow, conditions resolve against real records (literacy training, signed DPAs), and a combination no rule covers answers 'no policy covers this' — never a silent yes.

Every REST or MCP answer lands in an append-only decision log. When your DPO or an auditor asks 'how do you know the policy is followed?', the answer is a log of every check — not an assurance.

Enforcement stays yours. The open-source Clearance Enforcer plugins (LiteLLM hook, OpenClaw package, n8n node — MIT) call the API from inside your own stack, verify signed policy bundles offline, and fail open with a flagged receipt or closed, per your plan and configuration. Fronterio is never in the data plane.

Policy as a document vs. policy as infrastructure

Policy as a document

  • The routing matrix lives in a PDF the gateway has never read
  • Whether a call was policy-compliant is reconstructed after the fact, if at all
  • Every new runtime needs someone to remember and re-implement the rules
  • The auditor gets assurances, not evidence

Policy as infrastructure

  • Runtimes ask the PDP before the call and get the verdict plus approved alternatives
  • Every decision is logged append-only at the moment it happens
  • One policy, one API — gateways, agents and workflows all ask the same source
  • The evidence log answers the audit question before it's asked

What the Clearance API gives your platform team

The check endpoint

POST a model ref and data classes; get green/yellow/red with reason codes and the conditions that apply. 300 requests/min per key — built for request paths.

The recommend endpoint

When the model choice is still open: every approved register model ranked green-first for the data classes in play.

The decision log

Append-only, source-tagged (REST, MCP, preview, receipt), visible on the vendors screen and retained per your data-retention policy.

Signed policy bundles

Enforcer plugins fetch a signed snapshot of your policy and verify it offline — enforcement keeps working when the network doesn't.

Open-source enforcers

MIT-licensed plugins for LiteLLM, OpenClaw and n8n — auditable code in your stack, not a black box in ours.

Honest fail semantics

Fail-open with a loudly flagged receipt by default; org-configurable fail-closed on Business+. Either way, the behaviour is documented and evidenced.

How a call gets cleared

1

Write the policy once

The routing matrix on the vendors screen — data class × model or region × verdict, org-wide with per-agent overrides.

2

Point your runtime at the API

Drop in an enforcer plugin or call the REST endpoint directly with an API key scoped to policy decisions.

3

Ask before the call

The runtime asks, gets the verdict and alternatives, and enforces it in your own infrastructure.

4

Show the evidence

The append-only log on the vendors screen is the proof — every model call was policy-checked.

A model policy written as a PDF is a policy no gateway has ever read. The Clearance API makes the gateway ask before the call goes out. Every decision is logged, so “are the rules actually being followed” has an answer.

How the Model Clearance API works

Available from Business

The REST Clearance API is included from the Business plan; MCP tools ride the Enterprise MCP server.

BusinessEnterprise

Make your routing policy answerable

Document your vendors and rules once — then let every runtime ask before the call.

Model Clearance API — a policy decision point for your AI runtimes | Fronterio