ExAI brings an LLM-powered, in-game AI assistant to your Minecraft server. It quickly builds your server's knowledge base through automatic uploads and chat monitoring, and uses RAG to give accurate, server-specific answers. Everything happens inside a chest GUI — no command memorization required.

Project: https://github.com/Crossroadscodes/MC-ExAI

Features

• Chest GUI assistant — players talk to the AI through a clean menu, no chat spam.
• Public chat response — auto-replies to questions in public chat via keyword triggers (configurable cooldown); stays silent when no grounded answer is available to avoid spam.
• Auto knowledge collection — watches the "ask → answer → thanks" flow in public chat and automatically queues AI‑vetted Q&A pairs for OP review.
• Player knowledge submission — players submit Q&A using a written book → AI pre‑filters junk → OP approves/rejects in GUI → approved entries go into the vector store.
• Reward system — Vault currency + custom items granted upon approval.
• RAG semantic retrieval — uses DashScope text-embedding-v3 (1024‑dim) with adjustable similarity threshold.
• OP tools — chat history viewer, knowledge base editor, one‑click promotion of chat logs to knowledge.
• Flexible storage — switch between MySQL and pure local YAML via storage.type; zero database setup for small servers.
• LLM backend — OpenAI‑compatible API (default Aliyun DashScope/Qwen, easily swappable for OpenAI, DeepSeek, etc.).
• Low cost — achieved via configurable cooldown settings and output truncation to minimize AI calls.

Screenshots
Main menu — Start a chat
Hover the player head and click to begin a single-turn dialogue with the assistant.

Main menu — Single-turn chat notice
The paper item shows usage info and the current cooldown.

Main menu — Upload knowledge
Click the book to receive a writable book; fill in Q: ... A: ... and close it to submit for review.

Main menu — Reviewer panel (OP)
Open the per-page review GUI. Left-click to approve (writes to the knowledge base and pays rewards), right-click to reject.

Main menu — Chat history (OP)
Browse player questions and AI replies; left-click to edit and promote a record into the knowledge base, right-click to delete.

Main menu — Knowledge base (OP)
View and edit every accepted entry. Left-click to edit, right-click to delete; the knowledge base auto-reloads.

Public chat in action
Type a question with a configured keyword (?, how, what, etc.) and the assistant replies on broadcast.

Commands & Permissions
Command Description
/exai opengui Open the assistant GUI
/exai question <player> <q> Send a question to a player (OP)
/exai reload Reload config (OP)
/exai help Show help
Permission node: exai.op (knowledge review).

Quick Setup

1. Drop ExAI.jar into plugins/.
2. Start the server once → edit plugins/ExAI/config.yml:
   • Choose storage.type: yml (default, no DB) or mysql.
   • If mysql, fill in MySQL credentials.
   • Fill in your LLM API key (llm.apiKey).
   • Adjust assistant name / language / rewards.
3. (Optional) Edit plugins/ExAI/knowledge.yml to seed your knowledge base.
4. /exai reload.

Storage modes

• yml (default)
• mysql

Requirements

• Spigot/Paper 1.12.2 or compatible
• Java 8+
• MySQL 5.7+ — only if storage.type: mysql
• Vault — optional, for currency rewards
• An OpenAI-compatible LLM endpoint + embedding endpoint