Skip to main content

Architecture

This section explains how DIAL is built: its services, how requests move through them, what you deploy and operate, and where the limits are. It is for architects assessing DIAL for their organization. It assumes you have read What is DIAL; installation steps are in Operating DIAL.

In this section​

  • Architecture highlights — the services around DIAL Core, what each is responsible for, and where your own services plug in.
  • Request flow — the runtime paths for models, applications, MCP Toolsets, and routes, and how interceptors process them.
  • Deployment profiles — which services to adopt for the capabilities you need, and the infrastructure and ownership behind each choice.
  • State and data lifecycle — how DIAL Core separates service instances from stored resources, for planning persistence, retention, and recovery.
  • Scalability and performance — which components keep state, how each one scales, and what load tests showed.
  • Application hosting and lifecycle — who hosts and runs each application type, from authoring to operation.
  • AI API overview — the Unified API, the other supported AI APIs, passthrough and translation, and multimodal models.
  • Trade-offs and limitations — the architectural trade-offs, the limitations that can rule out a deployment, and the cost boundary.
  • DIAL Stack — reference for the technologies and infrastructure behind DIAL.

Suggested reading path​

The pages build on each other. Read Architecture highlights and Request flow first, then Deployment profiles before State and data lifecycle and Scalability and performance. The remaining pages can be read in any order.

To go straight to a specific decision, use the table in Evaluate your architecture.