Checkpoint Docs
Govern (KYA-OS)

Govern

Deploy KYA-OS servers for AI agent governance with delegations, proofs, and consent flows

What is Govern?

Checkpoint Govern enables you to deploy and manage KYA-OS ("Know Your Agent — Operating System"; see the glossary) servers for comprehensive AI agent governance. While Detect identifies agents and Enforce blocks or redirects them, Govern provides a framework for authorizing agents — letting verified AI agents interact with your resources under controlled conditions.

Govern implements a zero-trust model: every AI agent request must carry a cryptographic proof of identity and authorization before accessing protected resources.

How It Works

AI Agent → KYA-OS Proof → Your Server → Checkpoint Verification → Protected Resource

                                    Dashboard (audit trail)
  1. An AI agent obtains a delegation (authorization grant) through an OAuth flow
  2. The agent creates a cryptographic proof for each request, referencing the delegation
  3. Your server uses @kya-os/bouncer-middleware to verify the proof with Checkpoint
  4. If valid, the request proceeds with verified agent identity attached
  5. All activity is logged to the Checkpoint dashboard for audit

Key Concepts

Agent DIDs

Every AI agent is identified by a DID (Decentralized Identifier), such as did:key:z6Mk.... This provides a stable, verifiable identity for agents across sessions and services.

Delegations

A delegation is an authorization grant from a user or service to an AI agent. It specifies what the agent can do (scopes), for how long, and under what constraints. Think of delegations as OAuth tokens for AI agents.

Learn more about delegations

Proofs

A proof is a cryptographic assertion attached to each request. It contains the agent's DID, the target audience, a nonce, and a reference to the delegation that authorizes the request.

Learn more about proof verification

Scopes

Scopes define fine-grained permissions. For example, files:read allows reading files while payment:create allows initiating payments. Each delegation grants specific scopes, and each protected endpoint can require specific scopes.

Verifiable Credentials

Checkpoint issues and verifies verifiable credentials that attest to agent identity and permissions. These are signed by Checkpoint's infrastructure and can be independently verified.

Deployment Options

Checkpoint supports three ways to run a KYA-OS server:

ModelDescriptionInfrastructure
Dashboard DeployOne-click setup with GitHub repo + Cloudflare WorkerYour GitHub + Cloudflare account
MoltiManaged hosting — Checkpoint runs the agent, no repo or Cloudflare account requiredCheckpoint's infrastructure
Self-Host (BYOK — bring-your-own-key)Full control with KYA-OS packagesYour infrastructure

Deploy a KYA-OS Server →

Migrating an Existing MCP Server

Already have a standard MCP server? Add KYA-OS identity and authorization without rewriting your server.

Migration Guide →

Authentication Methods

When an agent requests access, the user must authenticate. Checkpoint supports multiple methods:

MethodDescription
Consent OnlySimple approve/deny — no external auth required
OAuthSign in with GitHub, Google, Microsoft, and more
Custom ProviderConnect your own authentication backend
CredentialsUsername/password or API key authentication

Compare authentication methods →

Prerequisites

  • A Checkpoint account with a project
  • Your project's API key and Project ID — see Credentials for where to find them
  • A Node.js/Express server (or equivalent) to add the middleware to

Quick Start

1. Install the middleware

npm install @kya-os/bouncer-middleware

2. Protect an endpoint

import express from 'express';
import { createBouncerMiddleware } from '@kya-os/bouncer-middleware';

const app = express();
app.use(express.json());

app.post(
  '/api/files',
  createBouncerMiddleware({
    apiKey: process.env.AGENTSHIELD_API_KEY!,
    projectId: process.env.AGENTSHIELD_PROJECT_ID!,
    requiredScopes: ['files:write'],
    reputationThreshold: 60,
  }),
  (req, res) => {
    // Access verified agent info
    const { agentDid, scopes, reputation } = req.bouncer;
    res.json({ message: 'File created', agent: agentDid });
  }
);

Two env names, one key: AGENTSHIELD_API_KEY is the KYA-OS worker/bouncer naming convention for the same dashboard API key the Checkpoint SDKs read as CHECKPOINT_API_KEY. The bouncer middleware reads no environment variables itself — you pass values explicitly, as above.

3. Configure tools in the dashboard

Navigate to Policy → Auth (/dashboard/{orgId}/{projectId}/policy/auth) to define which tools require delegations and what scopes they need: create an auth method — OAuth or credentials when the user has to sign in, or consent-only when they just have to approve — plus the scopes it grants, then assign it as the owner of each tool it should protect. Auth methods are reusable, so one can own many tools.

Not every auth method involves signing in. Consent-only methods just ask the user to approve the scopes an agent is requesting, which is the right shape for delegation without an external identity provider. See Auth Methods for the comparison. The dashboard currently labels this section Sign-in methods; the underlying concept is the broader auth method.

Features

FeatureDescription
DeploymentDeploy KYA-OS servers via dashboard, Molti, or self-host
MoltiManaged agent hosting on Fly.io with lifecycle management
MigrationAdd KYA-OS to an existing MCP server
KYA-OS EnforcementBlock agents at the edge until they present a signed identity
Auth MethodsOAuth, Consent Only, Credentials, Custom Providers
DelegationsCreate and manage authorization grants for AI agents
ProofsVerify cryptographic proofs on every request
OAuthStandard OAuth flows for agent authorization
GitHub ConnectionsLet users connect GitHub so gateway tool calls carry their token
Tool ProtectionPer-tool scope requirements and access control
Consent FlowsUser consent UI for agent authorization

Govern vs Enforce

GovernEnforce
PurposeAuthorize known agentsBlock unknown agents
ModelAllow-list with cryptographic verificationDetect and apply policies
Use caseMCP server developers, API providersWebsite and app developers
Integration@kya-os/bouncer-middlewareGateway (DNS) or Middleware (code)

Govern and Enforce are complementary. Use Enforce to protect your web application from unwanted bots, and Govern to authorize legitimate AI agents that need to interact with your APIs.

Dashboard

Manage Govern features from the Checkpoint dashboard:

  • Delegations — View and revoke active delegations
  • Proofs — Monitor verification activity and audit agent access
  • Policy → Auth — Configure auth methods and providers, customize the consent screens agents see, and assign per-tool protection
  • Gateway → Connections — Let users link upstream OAuth accounts (e.g. GitHub) so gateway tool calls carry their credentials — a separate concept from delegations; see GitHub Connections
  • Control Access (legacy surface) — Tool discovery/removal and a few consent extras remain at /dashboard/{orgId}/{projectId}/control-access until the new Access surface lands — see Control Access

Checkpoint Govern is part of a larger KYA-OS ecosystem. These external projects provide additional context and tooling:

ProjectURLWhat it provides
KYA-OS reference impl.npmjs.com/package/@kya-os/mcp@kya-os/mcp — the protocol reference implementation: delegation, proofs, sessions, and audit records
KYA-OS protocol speckya-os.orgFull protocol specification covering credential models, verification protocol, audit layer, and reputation
KYA-OS for MCPkya-os.org/mcpHow KYA-OS binds onto the Model Context Protocol
KnowThat.aiknowthat.aiKYA-OS agent registry — browse agents, check reputation scores, claim an agent profile

Next Steps