Self-documenting portal
AI4IA — a governed, multi-model agentic chat platform on Azure
Governed multimodal and agent chat for enterprise knowledge work, built as an Azure capability showcase. A Next.js web app, a FastAPI governance backend, a SimpleL7Proxy model gateway, and Azure infrastructure as Bicep — all in one monorepo, all deployed through azd. This portal explains what it is, how it fits together, what it runs on, and what the latest snapshot reported.
What it is
The web frontend calls the FastAPI backend; the backend owns auth, session state, tools, memory, document/library access, usage metering, and all model routing. Compatible HTTP/SSE calls go SimpleL7Proxy -> APIM -> Foundry; Voice Live uses the FastAPI relay -> APIM path.
🎯 Purpose & principles
Gateway-first: HTTP/SSE uses SimpleL7Proxy -> APIM; realtime uses the relay -> APIM.
Catalog-driven: infra/models.json is the deployment source of truth.
Feature-gated: advanced surfaces fail closed and default safe.
Identity-first: Azure data planes use managed identity; scoped hop keys remain app secrets.
Cosmos is canonical: sessions, messages, usage, user agents/workflows, MCP server records, document manifests, and memory text/vectors. Search indexes, document chunks, and parsed artifacts are rebuildable.
🌍 Live environment
Capabilities
Dated observed production posture, separate from template defaults. Advanced capabilities are individually gated; repository and azd variables can override checked-in defaults. CI validates the defaults but does not query live Azure.
Technology
One monorepo spanning frontend, backend, gateway and infrastructure.
| Layer | Technology | Hosted on |
|---|
Explore
🟢 Status
Timestamped deployment snapshot: Azure resource health and public endpoint reachability.
🧭 Architecture
Diagrams of the request lifecycle, deployment topology, identity and MCP planes.
☁️ Services
Every Azure service AI4IA deploys, what it does and the IaC that creates it.
📋 Requirements
Infrastructure, permissions/RBAC and package dependencies.
📚 Docs
Curated index into the repo's Markdown: overview, architecture, configuration and runbooks.
🚀 Deploy to Azure
Use the guided workflow for quota, identity, infrastructure, images, and verification.