Azure AI Gateway Tier: Preview Setup, Policies, MCP, and Token Monitoring
A source-based technical evaluation of the dedicated AI Gateway preview tier, including measured policy behavior and application-level token reporting.
AZURE / APPLIED ENGINEERING
Practical field notes on AI gateways, Kubernetes, and App Service. Procedures, measured results, and the caveats that matter.
English guides
Parameterized examples
Source-derived downloads
THE COLLECTION
10 topics
A source-based technical evaluation of the dedicated AI Gateway preview tier, including measured policy behavior and application-level token reporting.
A consolidated implementation and test record for Azure-hosted Codex, including identity, load balancing, mid-stream 429 behavior, and PTU/PAYG sizing.
Configure existing APIM Standard v2 and Foundry resources with private backend access, bounded 429 retry, and privacy-conscious Application Insights diagnostics.
Measured translation latency, token output, migration options, and limitations from a six-model Azure OpenAI comparison.
Migrate an internal NGINX ingress to managed Gateway API while separating current operator-managed TLS from historical manual CSI guidance.
Enable ACR artifact streaming for eligible Linux node pools and measure cold-start improvement without cache bias.
Reproduce a measured Azure CNI Overlay migration and progressively enforce DNS, cross-namespace, and external-access policies.
Deploy an internal Envoy Gateway with independent data planes, safe draining, and explicit replacements for NGINX routing features.
Understand observed front-end scaling delays, distinguish configured capacity from actual instances, and plan capacity ahead of demand.
Explore plan-scoped Windows customization, app-scoped private endpoints, Bastion diagnostics, and the limitations observed during a preview PoC.
No topics match your search. Try another term or choose All topics.
Each topic brings related technical notes together in one page, with a section index, copyable examples, and companion downloads when available.
Environment identifiers are replaced with variables or placeholders. Original private screenshots, customer assessments, and unavailable local files are not distributed.