Selected assistant context
Ask Portal Architect Start with the job: check onboarding readiness or diagnose a live answer. Architect reads only this signed-in account and names the owning layer: instructions, knowledge, playbook, policy, platform/runtime, or retest.
Approval required for changes Website Channel
Choose the exact website and channel. Every Architect request remains bound to this selection.
Status Loading server-resolved context… Finish onboarding Find the first account-specific setup gap. Diagnose a live answer Classify bounded review evidence without treating a transcript as a business fact. Diagnose answer issues Use AI Chat Logs evidence to propose specific, human-reviewed improvements.
Architect will cite inspected Portal areas and state when an authority is unavailable.
Inspected configuration layers Loading knowledge, runtime, handoff, playbook, and policy status…
Draft · never live
Greeting and instruction change set Draft and test proposed values against the current published version. An authorised human must explicitly approve before publication.
No draft Proposed greeting Proposed assistant instructions Reason Risks (one per line) Create draft Validate revision Compare before / after Discard draft
Published values are live. Draft values are isolated.
No-send prompts (one per line) Run revision-bound no-send tests
No-send results bind to the current draft revision and create no customer conversation.
Denial reason Deny validated revision Approve current revision Publish approved revision Rollback Architect publication
Approval and publication controls appear only when your permission and the saved workflow state allow them.
Drafts never alter published configuration. Publish and rollback use version checks and authoritative saved-state readback.