Technology

The Architecture Behind the Universal Gateway

A deep dive into how Relay.one connects, governs, and traces your AI agent ecosystem with zero vendor lock-in.

Core Capabilities

Four Foundational Pillars

Relay Connect

Universal Protocol Translation

Seamlessly bridge REST APIs, Model Context Protocol (MCP), and Agent-to-Agent (A2A) communications. No custom wrappers or adapters required.

REST ↔ MCP translation
A2A protocol support
Zero-config routing
Framework agnostic

Relay Govern

Context-Aware Governance

Policy enforcement that understands prompts and context, not just packets. Intelligent PII redaction and compliance controls built-in.

Semantic policy engine
PII auto-redaction
Compliance templates
Custom rule builder

Relay Trace

Immutable Audit Trail

Every agent receives a UUID v7 identity. Every action is hash-chained for cryptographic audit integrity.

UUID v7 identity
Hash-chained logs
Tamper-proof records
Exportable audit

Relay Ledger

Usage Metering & FinOps

Track costs per agent, per team, per workflow. Turn AI from an unmanaged cost center into a measurable value stream.

Per-agent cost attribution
Budget alerts
Internal chargeback
Cross-org monetization

Architecture

Gateway Infrastructure

Self-hosted on your Kubernetes cluster. Zero vendor access. Complete control.

Ingress Controller

Always On

Load balancing & TLS termination

Policy Engine

Real-Time

Rule evaluation & enforcement

Audit Logger

Immutable

Hash-chained event recording

Provider Connectors

Multi-Cloud

Unified provider interface

Zero Trust Architecture

All components run within your infrastructure. Relay.one never sees your data, prompts, or agent communications. Complete self-sovereignty with no external dependencies.

Advanced Features

Enterprise-Grade Capabilities

Multi-Provider Routing

Route requests across multiple AI providers—Anthropic, OpenAI, Google Gemini, and local models—with intelligent load balancing and failover.

Real-Time Observability

Monitor gateway health, track latency percentiles, and view real-time activity streams across your entire agent fleet.

Cost Intelligence

Track spending across providers, set budget alerts, and optimize costs with detailed analytics and usage breakdowns.

Gateway Health Metrics

Monitor ingress controllers, policy engines, audit loggers, and provider connectors with sub-second health checks.

Policy Violation Detection

Real-time detection and blocking of policy violations with detailed logging for compliance and security review.

Human-in-the-Loop (HITL)

Queue high-risk agent actions for human approval while maintaining agent synchronization—no timeouts, no crashes.

Production Readiness

The Green Light Criteria

Six requirements that transform "Black Box Fear" into production confidence.

Identity

Know Who Is Acting

Cryptographic agent identity for every transaction

Least Privilege

Agents See Only What They Need

Scoped, minimal access by default

Policy Enforcement

Safety Outside the Model

Deterministic rules, not probabilistic prompts

Evidence

Ground Truth, Not Agent Storytelling

Immutable, hash-chained audit logs

Containment

Stop and Limit Blast Radius

Revoke, pause, and recover instantly

Cost Governance

Know What You're Buying

Per-agent metering and budget controls

Security Framework

The 6-Layer Trust Stack

Zero-trust security applied to AI agents. Every layer must pass.

Containment

Revoke, rate-limit, and recover from failures

Evidence

Aircraft black box logging for every action

Approvals

Surgical human sign-off for high-risk operations

Policy Enforcement

Deterministic rules that cannot be bypassed

Least Privilege

Minimum access required for the specific job

Identity

Cryptographic agent identity for every transaction

Compliance

Regulatory Alignment

Built to support SEC, NIST AI RMF, and EU AI Act requirements out of the box.

SEC Cyber Disclosure

Item 1.05 Ready

Ground-truth audit logs provide the evidence trail required for materiality assessments and incident disclosure.

NIST AI RMF 1.0

Govern · Map · Measure · Manage

Control plane architecture maps directly to NIST's four core functions for AI risk management.

EU AI Act

Risk-Based Classification

Policy enforcement and evidence production designed for high-risk AI system compliance.

Discovery

Agent Registry & Discovery

Know what agents and tools exist across your organization, partners, and the ecosystem.

Internal Registry

Catalog all agents and tools in your org. Know what exists, what it does, and who owns it.

Federation & Peering

Let trusted partners' agents communicate without exposing public endpoints. Controlled cross-org discovery.

Public Marketplace

Discover and publish agents for B2B monetization, investor interfaces, or public data services.

Economics

Usage Metering & Billing

Track costs, allocate spend, and monetize agent interactions—internally and externally.

Internal Showback

Per-agent and per-team cost attribution
Workflow-level cost breakdowns
Budget alerts and spend caps

Cross-Org Monetization

Per-call pricing and metering hooks
Prepaid credits and post-paid billing
Integration with Stripe, x402, and AP2

Safety

Staging & Simulation

Test agents safely before they touch production. Shadow mode, replay, and diff.

Shadow Mode

Agent proposes actions, platform logs and scores them—but no real changes happen. Safe testing at scale.

Replay Testing

Feed historical scenarios to see what the agent would have done. Validate before deploying.

Version Diffing

Compare agent v1 vs v2 behavior side-by-side. Know exactly what changes before rollout.

Operations

Operations Console

Human UX for operating agent infrastructure. Live traffic, incidents, and policy editing.

Live Traffic View

Top agents by cost, errors, and policy violations. Real-time drill-down.

Incident Response

Pause agents, revoke certs, force approval gates—all from one console.

Policy Editor

Non-engineer friendly UI for security and compliance teams to write rules.

Technical Moat

The Non-Blocking Interface

The Problem

Traditional architectures force agents to wait synchronously for human approvals. This causes timeouts, dropped connections, lost state, and requires manual intervention to restart.

The Solution

Relay's Smart Proxy holds the connection state, queues the approval request, and replays the response when authorized—all without the agent ever knowing it waited.

How It Works

Request

Agent initiates action

Hold

Relay pauses connection

Approve

Human authorizes

Replay

Success returned

Adoption Journey

The Path to Agentic Enterprise

A 5-stage maturity model: from shadow AI to optimized agent economy.

Shadow

Pilot Purgatory

Agents exist but are invisible. No inventory, no governance, no cost tracking.

Visibility

The Iron Gate

Inventory agent activity, map tool access, expose shadow AI. Create the baseline.

Guardrails

The Safe Harbor

Activate governance for high-value workflows. HITL for sensitive operations.

Approvals

Evidence & Audit

Standardized governance. Teams build agents with built-in compliance.

Optimize

The Agent Economy

FinOps for AI. Measure usage, allocate costs, reduce waste, optimize routing.

Ready to Deploy?

Get started with Relay.one on your infrastructure today.