premium banner

AI-Player

A minecraft mod which aims to add a "second player" into the game which will actually be intelligent.

Description

Read this section please.

This project so far is the result of thousands of hours of endless reasearch, trials and errors, and just the simple goal of eliminating loneliness from minecraft as much as possible. If you liked my work, please consider donating so that I can continue to work on this project in peace and can actually prove to my parents that my work is making a difference. (Also not having to ask for pocket money from mom).

Just know that I won't ever give up on this project.

Buy Me A Coffee

Paypal

https://paypal.me/shasankp000

Mod description

A minecraft mod which aims to add a "second player" into the game which will actually be intelligent.

Ever felt lonely while playing minecraft alone during that two-week phase? Well, this mod aims to solve that problem of loneliness, not just catering to this particular use case, but even (hopefully in the future) to play in multiplayer servers as well.

Please note that this is not some sort of a commercialised AI product. This project is just a solution to a problem many Minecraft players have faced and do continue to face.

I had to add that statement up there to prevent misunderstandings.

What does this mod do?

  This mod so far adds in a second spawnable player(or bot) via the carpet mod, and you can talk to it through a large language model via ollama completely offline on your pc and so far I have integrated reinforcement learning for the bot to interact and learn from it's environment in the context of "reflex actions".

This mod relies on the internal code of the Carpet mod, please star the repository of the mod: https://github.com/gnembon/fabric-carpet (Giving credit where it's due)

This mod also relies on the ollama4j project. https://github.com/amithkoujalgi/ollama4j

Changelog v1.0.5.1-release+1.20.6-bugfix-2

  • Fixed a lot of bugs that got overlooked in the previous testing phase.

Changelog v1.0.5.1-release+1.20.6-bugfix

I realized that many of the commitments made in the previous announcement were too ambitious to implement in a single update. To keep development smooth, this release is the first part of a two-part update.

This update focuses on core system rewrites and better AI decision-making, laying the foundation for the next wave of features.

Also support for versions below 1.20.6 has been dropped due to codebase changes that I simply can't handle migrating by myself. However others are free to port to lower versions.

Upcoming second part will have the update in 1.20.6 as the final update for 1.20.6 and also will have a new update to directly 1.21.6 where hence the future versions will continue onwards.

What's New in 1.0.5.1

Revamped NLP System

  • Fully redesigned Natural Language Processing (NLP) — no more "I couldn’t understand you."
  • This is a new and experimental system I’ve been designing and rigorously testing over the last month.
  • Results are promising, but not yet up to my personal standards — expect further refinements in future updates.

Rewritten RAG & Database System (with Web Search)

  • New Retrieval-Augmented Generation (RAG) system integrated with a database and web search.
  • The AI now provides accurate factual information about Minecraft, drastically reducing hallucinations.
  • Supported search providers:
    • Gemini API
    • Serper API
    • Brave Search API (in development, will push this to the next patch instead)

Meta-Decision Layer

  • Added a task chaining system:
    • You give a high-level instruction → the bot automatically breaks it into smaller tasks → executes step by step.

Current Supported Tasks:

  • Go to a location
  • Go to a location and mine resources
  • Detect nearby blocks & entities
  • Report stats (health, oxygen, hunger, etc.)

API Integrations

AI Player now supports multiple LLM providers (configurable via API keys):

  • OpenAI
  • Anthropic Claude
  • Google Gemini
  • xAI Grok

Web Search Tool

If you select the Gemini Search as the web search tool for the LLM, it will use the API key you have set as your LLM provider in the settings.json5 file automatically.

For https://serper.dev/ search, get an api key from serper.dev and then navigate to the config folder in game, open the ai_search_config.json and put the key:

Web search json file contents

(Note: I couldn’t test all of these myself except the Gemini API since API keys are costly, but the integrations are ready.)

Bugfixes

  • Fixed bug where JVM arguments were not being read.
  • Removed owo-lib. AI-Player now uses an in-house config system.
  • Fixed API keys saving issues.
  • Added a new Launcher Detection System. Since Modrinth launcher was conflicting by it's own variables path system so the QTable was not being loaded. Supports: Vanilla MC Launcher, Modrinth App, MultiMC, Prism Launcher, Curseforge launcher, ATLauncher, and even unknown launchers which would be unsupported by default, assuming they follow the vanilla mc launcher's path schemes.
  • Revamped the Config Manager UI with a responsive UI along with a search option for providers with a lot of models (like gemini).

Config Manager New UI with search options and responsive UI

Development Notes

  • While this update may look small on the surface, designing the systems, writing the code, and debugging took a huge amount of time.
  • On top of this, I’ve picked up more freelance contracts and need to focus on my final-year project.
  • Updates will continue — just at a slower pace.

Coming Soon in Part 2

Here’s what’s planned for the next patch:

  • Combat & Survival Enhancements

    • Bot uses weapons (including ranged) to fend off mobs.
    • Reflex module upgrades.
    • More natural world interactions (e.g., sleeping at night).

  • Improved Path Tracer

    • Smarter navigation through water and complex terrain.

  • Self-Goal Assignment System

    • Bot assigns itself goals like a real player.
    • Will initiate conversations with players and move autonomously.

  • Mood System (design phase)

    • Adds emotional context and varied behavior.

  • Player2 Integration

    • Highly requested — this will be the first major feature of the second update.

Upcoming changes (some of them might be seen in the second patch).

  1. Switch to Deep-Q learning instead of traditonal q-learning (TLDR: use a neural network instead of a table)
  2. Create custom movement code for the bot for precise movement instead of carpet's server sided movement code.
  3. Implement human consciousness level reasoning??? (to some degree maybe) (BIG MAYBE)

Sample video of the mod (as of the version 1.0.3-alpha-1)

Changelog video (as of the current version 1.0.3-alpha-2)

better and faster reflex actions, reduced training time.

environmental reaction

Youtube tutorial soon

Installation & Setup Guide

Step 1: Install Java 21

This project requires Java 21 to run.

  • Download Java 21 from BellSoft JDK 21 LTS.

  • On Windows: click Download MSI and complete installation.

  • On Linux: install via your package manager, e.g.

  sudo apt install openjdk-21-jdk

Step 2: Install Ollama

Go to

https://ollama.com/

Download based on your operating system.

After installation, run ollama from your desktop. This will launch the ollama server.

This can be accessed in your system tray

image

Step 3: Configure Ollama Models

Open your command line (CMD/Terminal) and run:

ollama pull nomic-embed-text   # Embedding model (required)
ollama pull qwen3:8b           # Preferred language model (used in testing)

For previous users:

ollama rm gemma2               # Remove old model if installed
ollama rm llama2               # Remove old model if installed
ollama rm llama3.2             # Remove old model if installed

If you’ve run the mod before, delete old config:

.minecraft/config/settings.json5

✅ Make sure the Ollama server is running before starting the game.

Step 4: Download Dependencies

Ensure that all required dependencies (including Carpet mod) are installed.

Step 5: JVM Argument Setup

Before launching the game, you must provide the following JVM argument:

-Daiplayer.llmMode=<provider>

Examples:

  • Local Ollama (default):
  -Daiplayer.llmMode=ollama
  • OpenAI:
  -Daiplayer.llmMode=openai
  • Anthropic Claude:
  -Daiplayer.llmMode=claude
  • Google Gemini:
  -Daiplayer.llmMode=gemini
  • xAI Grok:
  -Daiplayer.llmMode=grok

Step 6: Configure API Keys

Inside the Config Manager, there is now a dedicated section for entering API keys for all supported providers.

  • For local models, you can select your preferred Ollama LLM.
  • For hosted providers, enter your API keys accordingly.

⚠️ Note: Ollama must always be running since the embedding model (nomic-embed-text) is local and required.

Step 7: Launch the Game

  • Start Minecraft.
  • Before loading up a world (if you are on singleplayer) once the world loading screen reaches 100% it might lag a bit (or not) due to an NLP model being loaded in game
  • Before loading up a world, please wait till you see all the folders in a structure with files like this: NLP Models directory structure
  • These files are for a one-time download only, so it's only for the first time on using this version.
  • Once in the game, if you used a different provider other than ollama in the JVM argument, use the config manager to select your desired LLM/provider and set their API keys.
  • Type the following command to spawn your bot:
/bot spawn <yourBotName> <training|play>
  • training → Spawns bot in training mode (does not connect to LLM).
  • play → Normal mode with full AI features.

Important Notes

  • It’s best to create a separate modpack for this mod, since Ollama must run in the background whenever you play.
  • Ollama itself uses minimal memory, so it shouldn’t affect performance significantly.

Web Search Tool

If you select the Gemini Search as the web search tool for the LLM, it will use the API key you have set as your LLM provider in the settings.json5 file automatically.

For https://serper.dev/ search, get an api key from serper.dev and then navigate to the config folder in game, open the ai_search_config.json and put the key:

Web search json file contents

Mod commands

/bot spawn <bot> <mode: training or play>, if you type anything else in the mode parameter you will get a message in chat showing the correct usage of this command

/bot use-key <W,S, A, D, LSHIFT, SPRINT, UNSNEAK, UNSPRINT> <bot>

/bot release-all-keys <bot> <botName>

/bot look <north, south, east, west>

/bot detectDangerZone // Detects lava pools and cliffs nearby

/bot getHotBarItems // returns a list of the items in it's hotbar

/bot getSelectedItem // gets the currently selected item

/bot getHungerLevel // gets it's hunger levels

/bot getOxygenLevel // gets the oxygen level of the bot

/bot equipArmor // gets the bot to put on the best armor in it's inventory

/bot removeArmor // removes the armor from the bot.