Architecture
Follow an inference request
Where your prompt travels and which details Sonnely keeps.
When you send a message in the console, your browser sends it over HTTPS to Sonnely. The Sonnely server reads the request in order to validate it, reserve credits, and forward it to NEAR AI Cloud.
The catalog uses models that NEAR AI lists under Private TEE. This tier describes the provider’s execution environment. The Sonnely server is a separate trust boundary. The standard Sonnely gateway does not provide client-to-model end-to-end encryption.
The account ledger stores the model ID, API key label, token counts, request status, and charge. Prompts and responses are excluded from the ledger. If you enable chat history, the console stores it in your browser.