AI Director
AI Director is a control layer for the owner, not autopilot. It reads company records, explains what needs a decision, and proposes an action. A human approves, dismisses, or waits.
The commercial “AI Director” module (KZT 9,900/month) opens chat, briefings, and the decision log. Recommendation kernel is in the platform; owner and admin decide on cards.
What it may do
- chat over company data, including voice;
- morning briefing, daily and weekly digests;
- recommendations and risks that need a decision;
- proposed actions after you confirm;
- decision log;
- review queue (
/ai/triageand related screens): filters, dismiss reason, reopen when underlying facts change, review SLA, bulk-dismiss of low/medium items only (cap 25, no bulk approve).
Command center: /ai-director. Chat: owner, admin, director. Decide on a recommendation: owner and admin only.
What it must not do
- approve, assign, or change severity without a human (except reopening a previously dismissed card when deal/task facts actually changed — a rule, not “the model decided”);
- message customers or start campaigns by itself;
- close deals, post payments, or file tax returns;
- call OpenAI or Anthropic: only Google Gemini and Groq are in the perimeter.
Deal summaries
The Sales assistant module builds a summary on the deal card. That is not the Director queue, and it still needs a human before a customer email.
Knowledge base
The RAG module attaches your files. Answers cite sources. Without it, the briefing does not read uploaded PDFs as a knowledge base.
Quotas
Director settings show call counts and an estimated cost. A soft ~80% threshold and a hard daily cap stop new model calls rather than continuing silently.