AI connectivity for your own product
A stateless backend model adapter, ready to connect to the workflow you build.
On this page
Connect a model without imposing a chat product
The repository includes a Chat Completions-compatible text adapter in foundation/aix and configuration in the Go template. Your business service supplies messages, calls the adapter with the request context and handles the result. Provider credentials stay on the server.
Keep product decisions in product code
A chat application needs conversation ownership, message persistence, prompts and a retention policy. A generation tool needs task state, delivery and failure handling. Neither is supplied as a default customer product. Enabling the adapter does not create a public AI endpoint.
Reuse the credit service when your action charges credits, while defining success, retries and compensation in your own service. An external model request cannot be rolled back by a local database transaction.
Configure and verify
Choose a compatible endpoint and model, configure the server-side key, and inject the application HTTP client. Start with one authorized action and test provider failures before charging customers. Read the AI setup guide.