Architecture Overview

The Agentic Web Stack describes how modern AI agents can be exposed, discovered, integrated, and operated using open standards and emerging protocols.

The architecture treats agent-native systems as a layered web platform:

• Discovery tells agents what exists.
• Identity controls who or what may access it.
• APIs expose deterministic system capabilities.
• Agent protocols expose tools, resources, and task workflows.
• Memory gives agents durable context.
• Models provide reasoning and generation.
• Applications compose the lower layers into useful work.

Canonical Stack

Design Goals

• Keep the project vendor-neutral and standards-focused.
• Make every layer independently understandable.
• Show how human-readable documentation and machine-readable artifacts fit together.
• Keep protocol boundaries explicit so implementations remain interoperable.
• Avoid tying the architecture to one vendor, framework, or model provider.

Reference Artifacts

The architecture uses two kinds of artifacts:

• Human-readable explanations for architecture, security, operations, and technology choices.
• Machine-readable examples for Agent Cards, OpenAPI contracts, identity configuration, MCP tools, A2A messages, memory, model routing, and application workflows.

Next Step

Read the technology stack guide for common implementation technologies, then use the layer model for the full stack view and architecture variants for deployment shapes.

Then follow the End-to-End Demo, inspect Architecture Flows, and use Specifications for machine-readable examples.