An AI feature is more than one text response.
A real product may need a model catalogue, several provider accounts, streaming responses, embeddings, uploaded files, batch jobs, tool calls, usage records, and a way to stop or drain outstanding work.
If every screen talks directly to a provider, those decisions spread through the product. The source work brings them behind one Daptin-facing compatibility layer and connects them with stored credentials, deployments, files, usage, and runtime lifecycle.
The intended benefit is not a fashionable endpoint. It is keeping AI work inside the same product context and operating boundary as the application that requested it.