Skip to main content

Core Concepts

UAPK Gateway provides a policy enforcement layer between AI agents and the tools they use. This section explains the core concepts.

Architecture Overview

Request Flow

Every agent action request flows through the gateway:

Key Concepts

ConceptDescriptionLearn More
UAPK ManifestJSON document declaring agent identity, capabilities, and constraintsManifest
Capability TokenEd25519-signed JWT granting specific permissionsCapabilities
Policy DecisionALLOW, DENY, or ESCALATE based on evaluationDecisions
Approval WorkflowHuman review for escalated actionsApprovals
Audit LogHash-chained, signed InteractionRecordsLogs

Multi-Tenancy

UAPK Gateway supports multiple organizations with isolated data:

Each organization has:

  • Independent manifests and agents
  • Separate policies and approval workflows
  • Isolated audit logs
  • Distinct API keys and users

Security Model

UAPK Gateway implements defense-in-depth:

  1. Authentication: JWT tokens for users, API keys for agents
  2. Authorization: RBAC for operators, capability tokens for agents
  3. Policy Enforcement: Rules evaluated at runtime
  4. Audit Trail: Tamper-evident, cryptographically signed logs
  5. Key Management: Ed25519 keys for signing tokens and records

See Security for the full threat model.

Next Steps

UAPK Manifest

Learn how to declare agent identity and capabilities.

Capability Tokens

Understand fine-grained, time-limited permissions.

Policy Decisions

How the policy engine evaluates requests.

Approvals

Configure human-in-the-loop workflows.