Guides
Go-live checklist
From sandbox to production, deliberately.
Work through this list top to bottom. Plan for about two weeks from first install.
Before switching keys
- Every registered action has an accurate
description(it steers the model) and the right risk level. - Write and destructive actions have a clear
confirmTemplatenaming what will happen. - Knowledge sources are uploaded and show Ready in the console, which means their passages are indexed and retrievable — a source that is still Indexing cannot be found by the copilot. Test retrieval on the Knowledge page runs the same search a live turn runs, and shows the relevance score of every passage it returns.
- The persona (name, tone, instructions, guardrails) is set in the console.
- Your three most common user requests work end to end in sandbox, on a real device.
-
Rendel.reset()is called on logout, and identity verification is enabled if accounts matter (we provision the identity secret with you while the console control ships). - Theming reviewed in light and dark mode.
Switching
- Swap
rd_pk_test_forrd_pk_live_behind your feature flag. - Check the conversation cap and per-user rate limits on the app's Settings page, and email us if they need raising.
- Roll out gradually; read the Conversations page daily for the first week.
- Know the rotation window before you need it: rotating a key issues the new one immediately and keeps the old one working for seven days on live, one day on test. Ship the new key inside that window.
After launch
The Conversations page is your backlog: read the turns where the copilot could not help, and each one is either a missing knowledge passage or a missing action.