Assistant boundary
Useful, controlled, and reviewable
The AI surface is a product feature, not a loose chatbot.
Translate package failures into next steps users can actually follow.
Improve titles, descriptions, screenshots, compatibility notes, and submission evidence.
Answers should come from reviewed docs, release notes, and support decisions.
Usage limits, model settings, ratings, and group rules stay admin-controlled.
Telegram help can point users to the right topic without exposing private account data.
Billing, file access, root actions, and app-review wording remain auditable.
Knowledge system
Answers should come from CRYA-owned sources
The AI layer should cite and reuse reviewed CRYA material instead of inventing risky root, billing, or compatibility advice.
Package Doctor, QMG export, store submissions, billing, support evidence, and root safety should be first-class sources.
App, website, store, demo mirror, mail, and maintenance changes should become short, searchable knowledge entries.
Resolved support patterns can become reviewed answer snippets after private user data is removed.
Pack descriptions, compatibility notes, preview evidence, and store review decisions should improve recommendations.
Maintenance mode, heartbeat, known incidents, and service routes should inform safe retry guidance.
Privacy, terms, deletion, ads, billing, and root disclaimers should constrain what the assistant promises.
Review loop
From user question to safer answer
A good AI Observatory gives admins ways to inspect quality, costs, and risk before answers become official.
User asks from the website, app support path, or community topic.
Assistant searches reviewed CRYA docs, package reports, and status context first.
Reply stays short, user-facing, and avoids private internals or unsafe promises.
Users and moderators can flag helpful, wrong, risky, or outdated answers.
Useful answers become reviewed docs, snippets, or support templates.
Admin controls
Cost and safety stay visible
AI should feel helpful to users while still giving admins enough controls to prevent runaway spend or bad advice.
Monthly budget, reply cap, model choice, and fallback behavior.
Blocked topics for private internals, account secrets, and unsafe install claims.
Review queue for low-rated or policy-sensitive answers.
Source freshness checks for docs, app releases, store rules, and status incidents.
| Package help | Explain reports, desc.txt problems, frame order, dimensions, and export choices. |
|---|---|
| Creator help | Improve titles, descriptions, screenshots, compatibility labels, and store evidence. |
| Support routing | Collect evidence and send users to the correct private or public path. |
| Community help | Answer general questions without exposing account, billing, or admin data. |
AI routes
The Observatory has room to grow
These public routes should explain the AI product while the private queue, prompts, tokens, and moderation decisions stay inside Mission Control.
User-facing helper for package questions, creator copy, docs routing, and safe troubleshooting.
OpenSourcesKnowledgeReviewed source map for docs, releases, support decisions, store rules, and status signals.
OpenGuardrailsSafetyPolicy boundaries for root guidance, private data, billing, install risk, and platform review.
OpenAdminReview queueAdmin-reviewed answer quality, flags, cost controls, and publishable knowledge candidates.
Open