What happens to my prompts?
The credit service passes each request to the model that answers it. Its code never stores or logs the text of your prompts or the answers. Per request it records counts: tokens, credits, the model alias, the status and the timing. That is what your usage table is built from.
Your prompt does go to the model that answers it. With an AI agent, so does whatever it reads while working on your request: file contents and command output.
The credit service is open source: mcxVirsilipas on GitHub.
Keeping your key safe
- Anyone with your key can spend your credits. Keep it out of web folders, torrents, screenshots and shared files.
- On the seedbox, keep
~/.virsilipasKeyreadable only by you:chmod 600 ~/.virsilipasKey. - If it may have leaked, press Get a new key on the Virsilipas AI key page. The old key stops working at once.
Using an AI agent safely
- An agent can delete or overwrite anything your account can. Back up data you cannot replace first.
- Read each command before you approve it, and avoid fully automatic modes while the agent works on downloaded files or web pages: text inside them can carry instructions aimed at the agent.
- Point the agent at the directories it needs and keep it away from private files.
More on running agents on a seedbox: Using them safely on the wiki.
