Security and RBAC

We understand the risks associated with using LLM-based applications with business data, and we have implemented multiple safeguards to ensure compliance and security.

1. Admin-controlled Autopilot functions

   All Autopilot capabilities can be configured by admins through the AI Trust Layer. Admins can enable, disable, or modify LLM settings as needed.

   

2. Private, browser-local conversations

   Individual conversations persist only in the user’s local browser and are unique to that user. No other user can view these conversations, and they are not stored in the backend. Users may delete them at any time.

3. Governed access to traces

   Agent traces are governed at the AI Trust Layer. Trace summaries are generated on the fly by the LLM and are not stored outside the conversation. A user can view a trace summary only if they have access to the corresponding task.

4. Protection against inappropriate or harmful prompts

   NSFW, unrelated, or destructive prompts are explicitly restricted. Users are prevented from performing such actions.

5. Configurable tools and prompts

   All tools and prompt capabilities can be enabled or disabled at the Autopilot settings level.