This mode does not make the app autonomous. It generates a controlled LLM prompt with a defined mission, allowed inputs, permitted actions, hard boundaries, stop rules and a maximum number of reasoning cycles.

• Goal: what the agent is allowed to help with.
• Allowed inputs: which sources the agent may use.
• Permitted actions: what output types it may generate.
• Hard boundaries: safety limits, escalation rules and uncertainty handling.

Use it for structured clinical reasoning, documentation and safe next-step planning—not for uncontrolled external actions or final diagnosis replacement.