Your API session. Clear privacy boundaries.
A dedicated GPU instance for model inference. Save responses in your own client; account metadata and external copies have separate lifetimes.
Your instance is dedicated to you.
Your model runs in a GPU instance dedicated to you. Other customers do not use your instance.
No hosted chat history.
The API processes prompts and responses during your session. The application is designed not to store that content in its permanent database or content logs. It does not provide a hosted conversation history or file workspace.
This is not a promise of physical erasure across every infrastructure layer. Host, crash-log, backup and infrastructure-provider retention have not been independently audited end to end.
Keep responses in your own client.
Save any prompts or responses you need in your client. There is no hosted file upload or export service. Resource absence confirms cleanup through the provider API; it does not prove physical disk erasure or the absence of provider backups.
End revokes API access and stops customer billing. The service requests resource deletion and reports cleanup as pending until absence is confirmed. Closing the browser or canceling a request does not end the session.
The dashboard displays the funded deadline. API access stops at expiry; your own saved copies remain with you.
Your account stays.
Your account, password hash, API-key hash and masked key metadata, wallet, billing and session-lifecycle records remain after a session ends. Your API endpoint and account key remain available for later sessions. Ending a session does not delete your account.
For security and rate limiting, we use your client IP address. Our request logs record only the client IP, request method, route category, status and duration. They exclude request bodies, URL query strings and credentials. These local logs are rotated and retained for up to 14 days.
The metadata schema has no prompt, response or uploaded-file records. Final retention periods and account-deletion procedures must be published before paid launch.
Copies outside the session.
Your own client or harness may save conversations and tool results. You control those copies and the external services your harness contacts.
Public HTTPS traffic, including API requests and responses, passes through Cloudflare. Its infrastructure processing and retention are separate from our application’s local logging controls.
Native tool calls are data returned by the model. Your harness decides whether to execute them; any external service it contacts has its own terms and retention practices.
How sessions work