Skip to main content
You can read this guide as a short introduction, follow one request in detail or jump to a specific reliability question. Each route below stays inside the guide; none requires access to an internal source archive.

​
I want to understand the idea

Begin with Why Gateway exists, then How Gateway works. Finish with Who owns what. After these pages, you should be able to explain why one shared execution core is useful, where model execution ends and where application responsibility begins.

​
I am building an application around Gateway

Read Your first request, Planning a compatible request and Restarting without repeating the work. For live output, add Following a live stream. For machine-consumed output, read When the answer must follow a schema. These are design explanations, not an invented SDK quickstart. Before implementing an integration, use the exact engineering contract and current interface for your deployment. The engineering reading map explains which topics need that deeper source.

​
I need to understand a failure

Start with When a confirmation is lost. Then read Attempts and retries and The point where an answer cannot be replaced. The key question is not just “which error occurred?” It is “what is known about the original operation, who still owns it and what action is still permitted?”

​
I operate or review the system

Read Ready to serve is more than running, Investigating a request and Recovery after a storage failure. Compare those intended controls with What has actually been demonstrated. A healthy process, a local test and a production recovery guarantee are different claims.

​
I am interested in cost or business work

Use Usage, provider cost and customer payment, The owners of a payment and A model call inside a larger task. These pages explain why a successful model answer cannot stand in for accepted work or a settled financial operation. The glossary defines terms in plain English. The FAQ answers the questions that most often cross several chapters.