LLM Connect

Add to collection...
Report an issue Ask a question

Description

⚠️ Important Setup Note 🔗

Add the following line to your minetest.conf for API connectivity:

secure.http_mods = llm_connect

About LLM Connect 🔗

LLM Connect turns Luanti into a Lua-first AI-assisted development and agent environment. It connects your server to OpenAI-compatible LLM endpoints, local or hosted, and provides integrated chat, Smart Lua IDE tooling, and modular agent skills.

Features 🔗

AI Chat 🔗

  • Talk to an LLM directly from inside Luanti.
  • Context-aware responses using relevant server and player state.
  • Configurable provider support through OpenAI-compatible endpoints.

Smart Lua IDE 🔗

  • In-game editor with code and output panels.

  • AI code generation, explanation and repair workflow.

  • Sandboxed execution for llm_dev.

  • Owner-level development workflow for llm_root.

  • World-backed snippet storage compatible with Luanti 5.16+.

  • Explicit cold reload management through the File Manager:

    • Run executes code transiently.
    • Save persists snippets without enabling startup execution.
    • Enable on Restart loads selected scripts during the next server/world startup.

  • Supports startup registration workflows for generated nodes, tools, entities and crafts.

Lua-first Agent and Skills 🔗

  • Dual-channel architecture: visible assistant replies plus structured Lua actions.

  • Centralized execution through the runtime executor.

  • Capability-separated permissions for chat, development, agents and root access.

  • Internal skill modules:

    • command_agent
    • mapgen_painter
    • node_printer_preview

  • External Lua-first skill mods can register through the skill registry and are validated at startup.

Supported API Providers 🔗

Any compatible OpenAI-style endpoint may work, including:

  • OpenAI API
  • Mistral API
  • Ollama
  • LocalAI
  • Open WebUI and compatible local gateways

Privileges 🔗

Privilege Access
llm AI chat
llm_dev Smart Lua IDE and sandboxed code execution
llm_agent Agent mode and skill tools
llm_root Full LLM Connect owner access

Quick Start 🔗

  1. Add secure.http_mods = llm_connect to minetest.conf.
  2. Enable the mod for your world.
  3. Configure your API in-game using /llm_config.
  4. Grant appropriate privileges, for example: /grant <name> llm_root.
  5. Open LLM Connect with /llm.

Reviews

Review

Do you recommend this mod?

  • No reviews, yet.

Releases

1.2.3

Download

2026-06-02 15:44 UTC

1.2.3 - some fixes and improvements 🔗

LLM Connect update 1.2.3

Improved IDE GUI: the left-side panel now doubles as the "Generate" input field

Automatic creation of IDE script storage on startup

Removed legacy code prefix requirement

1.2.0

Download

2026-05-29 19:36 UTC

1.2.0 - SKILL-BASED LLM AGENCY 🔗

LLM Connect 1.2.0

This is a major update from version 0.9.0.

New

  • Added a new skill-based agent mode for extending AI capabilities in a modular way.
  • Added new built-in tools for commands, map generation experiments and node-plan previews.
  • Added support for external skill mods.
  • Added a new persistent script workflow in the Smart Lua IDE:
    • Save Lua snippets inside the active world.
    • Choose which scripts should run when the world starts.
    • Use this to create persistent custom nodes, tools, entities and crafts from the IDE.

Changed

  • Rebuilt the internal AI execution system around Lua actions for more flexible agent behavior.
  • Reworked permissions:
    • llm_dev for IDE development and sandboxed testing.
    • llm_agent for agent features and skills.
    • llm_root for full owner/developer control.
  • Updated IDE persistence for compatibility with newer Luanti versions.
  • Replaced the former automatic startup execution of saved snippets with explicit Enable on Restart control in the file manager.

Removed / Deferred

  • Removed the old trusted mod-directory persistence approach.
  • True live hot reload is deferred for now; scripts that register nodes, tools or entities are activated on the next world/server start instead.

Migration notes

  • Workflows based on the older llm_worldedit permission should be migrated to the new llm_agent skill-based mode.
  • Scripts saved in earlier development builds are not automatically started; enable the scripts you want from the IDE file manager.
  • API access still requires:
secure.http_mods = llm_connect

0.9.0

Download

2026-03-10 22:01 UTC

MAJOR UPDATE - gui-chat, worldedit-agency and lua-ide 🔗

0.9.0 – Major Update

Complete rewrite of the development tooling. LLM Connect is no longer just a chat mod.

  • Smart Lua IDE – full in-game code editor with sandboxed execution, file manager and snippet persistence
  • Pre-Executor – static analysis before every run: syntax check, naming convention enforcement, required field validation
  • Auto-Fix Loop – LLM automatically corrects failed code; configurable iteration limit
  • Asset Picker – browse and inject Nodes, Items/Tools and Sounds directly into LLM context
  • API Reference Injection – optional Luanti API reference per session (slim / full)
  • WorldEdit Agency – AI-driven single-shot and loop building modes with WEA support
  • Material Picker – material selection UI for WorldEdit context
  • Comprehensive settings reference (/llm_config)
  • 29 response languages
  • Privilege system: llm, llm_dev, llm_worldedit, llm_root

0.7.8

Download

2025-12-05 00:10 UTC

0.7.8 - fall update 🔗

  • Revise README for Luanti mod updates
  • Bump LLM Connect Init version to 0.7.8
  • Modify mod.conf for version 0.7.8
  • Enhance settingtypes.txt with context options
  • Update system prompt for AI assistant description
  • Remove release version from mod.conf

2025-09-20

Download

2025-09-20 02:23 UTC

2025-09-20 🔗

  • Update license.txt
  • Update settingtypes.txt
All releases

Information

Provides

llm_connect

Dependencies

Required
No required dependencies

Information

Type
Mod
Technical Name
llm_connect
Languages
English
License
LGPL-3.0-or-later
Maintenance State
Actively Developed
Added
2025-09-11 09:43 UTC
Maintainers
Heisenberg