Responsible use

Keep the raw inference endpoint private and authenticated.
Treat generated text as untrusted; verify important claims independently.
Sandbox tool execution and require confirmation for destructive or external actions.
Apply least-privilege network and filesystem access.
Maintain rate limits, audit logs, incident response, and a rapid shutdown path.
Do not use the system to facilitate unlawful harm, evade legal obligations, or violate third-party rights.
Evaluate the assembled model for your own domain before deployment.