Skip to main content

04. Trust And Governance

Goal​

Seraph should become more autonomous only when the trust model gets stronger with it.

Design Direction​

The system needs:

  • explicit policy modes
  • audit trails for meaningful actions
  • scoped secret usage
  • approval paths for high-risk execution
  • clear separation between safe observation, planning, and privileged action

Core Principle​

If Seraph is going to act with more agency than a chatbot, the human needs clearer control surfaces than a chatbot.

Open Research Questions​

  • where should approvals live as native product UX rather than just API mechanics?
  • how should policy differ for native tools, MCP tools, workflows, and future channels?
  • how should Seraph ask for permission without destroying flow?
  • how should privileged execution be isolated from the planning layer?