Review, test and publish: how the assistant keeps you in control
Your assistant never changes live replies in one go. Every important change goes through a draft, a test and your "yes".
The four stages #
| Stage | Tool | What changes for fans |
|---|---|---|
| Draft | create_ai_profile, edit_ai_profile | nothing |
| Test | playground_reply | nothing: the Playground never sends to Telegram |
| Publish | publish_ai_profile | the new brain is live, but only if replies are on |
| Replies on | set_workspace_auto_reply | the AI answers her fans |
Preview, then apply #
The tools that matter work in two steps. First a preview: the assistant shows you what would change. Then apply, with the fresh hash or revision the preview gave. If someone changed the same thing in between, the apply is refused: nothing is overwritten, the assistant reads again and shows you the new preview.
What to check in the test replies #
- She answers in her language, with her tone.
- The prices match the ones you gave.
- A fan who says he is under 18 gets no sale and no flirting. See Minor or distress: when Lana stops.
- A fan in another language gets a sensible answer.
You can also open Playground in the Console and read the same test conversations.
Keep the last word #
- Publish agency AI is needed to create a creator's first AI Brain (
setup_workspace_aiwith apply), to publish a draft and to turn replies on. Without it, the assistant can edit and test drafts of a brain that already exists, never publish. - For a new creator, either keep the box ticked for the setup and untick it afterwards with Edit, or create and link her brain in the Console first, then let the assistant work on drafts. You then publish in AI Brain yourself.
- Replies stay off after
setup_workspace_ai. They go on only withset_workspace_auto_reply, after your "yes". - To stop everything at once, ask: "Turn off her AI replies." Or pause her in the Console.
Before the first real fan, run through the Go-live checklist.