Baritone's chat control prefix is # by default. In Impact, you can also use .b as a prefix. (for example, .b click instead of #click)

Baritone commands can also by default be typed in the chatbox. However if you make a typo, like typing "gola 10000 10000" instead of "goal" it goes into public chat, which is bad, so using # is suggested.

To disable direct chat control (with no prefix), turn off the chatControl setting. To disable chat control with the # prefix, turn off the prefixControl setting. In Impact, .b cannot be disabled. Be careful that you don't leave yourself with all control methods disabled (if you do, reset your settings by deleting the file minecraft/baritone/settings.txt and relaunching).

Lots of the commands have changed, BUT #help is improved vastly (its clickable! commands have tab completion! oh my!).

Try #help I promise it won't just send you back here =)

"wtf where is cleararea" -> look at #help sel

"wtf where is goto death, goto waypoint" -> look at #help wp

just look at #help lmao

Watch this showcase video!

Tutorial playlist

All of these commands may need a prefix before them, as above ^.

help

To toggle a boolean setting, just say its name in chat (for example saying allowBreak toggles whether Baritone will consider breaking blocks). For a numeric setting, say its name then the new value (like primaryTimeoutMS 250). It's case insensitive. To reset a setting to its default value, say acceptableThrowawayItems reset. To reset all settings, say reset. To see all settings that have been modified from their default values, say modified.

Commands in Baritone:

• thisway 1000 then path to go in the direction you're facing for a thousand blocks
• goal x y z or goal x z or goal y, then path to set a goal to a certain coordinate then path to it
• goto x y z or goto x z or goto y to go to a certain coordinate (in a single step, starts going immediately)
• goal to set the goal to your player's feet
• goal clear to clear the goal
• cancel or stop to stop everything, forcecancel is also an option
• goto portal or goto ender_chest or goto block_type to go to a block. (in Impact, .goto is an alias for .b goto for the most part)
• mine diamond_ore iron_ore to mine diamond ore or iron ore (turn on the setting legitMine to only mine ores that it can actually see. It will explore randomly around y=11 until it finds them.) An amount of blocks can also be specified, for example, mine 64 diamond_ore.
• click to click your destination on the screen. Right click path to on top of the block, left click to path into it (either at foot level or eye level), and left click and drag to select an area (#help sel to see what you can do with that selection).
• follow player playerName to follow a player. follow players to follow any players in range (combine with Kill Aura for a fun time). follow entities to follow any entities. follow entity pig to follow entities of a specific type.
• wp for waypoints. A "tag" is like "home" (created automatically on right clicking a bed) or "death" (created automatically on death) or "user" (has to be created manually). So you might want #wp save user coolbiome, then to set the goal #wp goal coolbiome then #path to path to it. For death, #wp goal death will list waypoints under the "death" tag (remember stuff is clickable!)
• build to build a schematic. build blah.schematic will load schematics/blah.schematic and build it with the origin being your player feet. build blah.schematic x y z to set the origin. Any of those can be relative to your player (~ 69 ~-420 would build at x=player x, y=69, z=player z-420).
• schematica to build the schematic that is currently open in schematica
• tunnel to dig and make a tunnel, 1x2. It will only deviate from the straight line if necessary such as to avoid lava. For a dumber tunnel that is really just cleararea, you can tunnel 3 2 100, to clear an area 3 high, 2 wide, and 100 deep.
• farm to automatically harvest, replant, or bone meal crops. Use farm <range> or farm <range> <waypoint> to limit the max distance from the starting point or a waypoint.
• axis to go to an axis or diagonal axis at y=120 (axisHeight is a configurable setting, defaults to 120).
• explore x z to explore the world from the origin of x,z. Leave out x and z to default to player feet. This will continually path towards the closest chunk to the origin that it's never seen before. explorefilter filter.json with optional invert can be used to load in a list of chunks to load.
• invert to invert the current goal and path. This gets as far away from it as possible, instead of as close as possible. For example, do goal then invert to run as far as possible from where you're standing at the start.
• come tells Baritone to head towards your camera, useful when freecam doesn't move your player position.
• blacklist will stop baritone from going to the closest block so it won't attempt to get to it.
• eta to get information about the estimated time until the next segment and the goal, be aware that the ETA to your goal is really unprecise.
• proc to view miscellaneous information about the process currently controlling Baritone.
• repack to re-cache the chunks around you.
• gc to call System.gc() which may free up some memory.
• render to fix glitched chunk rendering without having to reload all of them.
• reloadall to reload Baritone's world cache or saveall to save Baritone's world cache.
• find to search through Baritone's cache and attempt to find the location of the block.
• surface or top to tell Baritone to head towards the closest surface-like area, this can be the surface or highest available air space.
• version to get the version of Baritone you're running
• damn daniel

All the settings and documentation are here. If you find HTML easier to read than Javadoc, you can look here.

There are about a hundred settings, but here are some fun / interesting / important ones that you might want to look at changing in normal usage of Baritone. The documentation for each can be found at the above links.

• allowBreak
• allowSprint
• allowPlace
• allowParkour
• allowParkourPlace
• blockPlacementPenalty
• renderCachedChunks (and cachedChunksOpacity) <-- very fun but you need a beefy computer
• avoidance (avoidance of mobs / mob spawners)
• legitMine
• followRadius
• backfill (fill in tunnels behind you)
• buildInLayers
• buildRepeatDistance and buildRepeatDirection
• worldExploringChunkOffset
• acceptableThrowawayItems
• blocksToAvoidBreaking
• mineScanDroppedItems
• allowDiagonalAscend
•  
• LGPL 3.0 License

ReinforceMC is an original project started by topandrey14 / topandrey15 in May 2025. The name "ReinforceMC" is associated with this project only. Any unauthorized use of the name in commercial or misleading projects is strictly prohibited.

ReinforceMC 🧠

A Minecraft mod that implements reinforcement learning capabilities, allowing neural networks to learn and play Minecraft through configurable reward systems and input processing.

Features ✨

🤖 DeepSeek + Baritone integration released!

• Message DeepSeek in your own Telegram bot, and it will perform actions in Minecraft using Baritone!

🎯 Custom Reinforcement learning Neural Network Training

• Start, stop, pause, and resume training sessions
• Real-time performance monitoring with reward tracking
• Emergency stop functionality for immediate training halt
• Model saving and loading capabilities

🏆 Configurable Reward System

• Set custom rewards for various in-game actions:
  • ⚔️ Combat: Kill players/mobs, take damage, heal
  • 🏗️ Building: Break/place blocks, craft items
  • 🚶 Exploration: Movement, sprinting, jumping, sneaking
  • 📦 Items: Pickup, drop, use, enchant items
  • 👤 Character: Gain XP, level changes, eat food

📊 Input Configuration

• 📷 Screenshot Input: Configurable resolution for visual learning
• 📍 Other Player Coordinates: Position and movement data
• 📈 Your Character Statistics: Health, hunger, XP, inventory information
• 👥 Nearby Player Tracking: Detect and track other players

📋 Reward Template System

• Save and load reward configurations
• Export templates to JSON format
• Create custom training scenarios
• Share configurations between sessions

🖥️ User Interface

• Comprehensive GUI for all configuration options
• Real-time training status and statistics
• Performance monitor with detailed metrics
• Available keys editor for input customization

Installation 🔧

📋 Requirements

• 🎮 Minecraft 1.16.5
• ⚡ Minecraft Forge 36.2.42 or compatible version
• ☕ Java 8 or higher
• 🖥️ GPU Support: NVIDIA GPU with CUDA 12.1 support (optional, falls back to CPU, but may be slower)
• 💾 Memory: At least 4GB RAM allocated to Minecraft

🚀 Setup

• The mod will automatically download required neural network libraries on first run, it may take up to 5 minutes for Minecraft to firstly load the mod. It will never happen again.

Usage 🎮

🌟 Getting Started

1. Open the ReinforceMC GUI using the configured keybind (default: check key bindings menu)
2. Configure your input settings (screenshot resolution, enabled inputs)
3. Set up reward parameters for desired behaviors
4. Configure available keys the AI can use
5. Start training and monitor progress

⌨️ Key Controls

• 🖥️ Open GUI: Configurable keybind
• 🔄 Toggle Training: Quick start/stop training
• 🛑 Emergency Stop: F1 (immediate pause/resume)

⚙️ Configuration

📥 Input Types

• 📷📍 Screenshot + Coordinates: Full visual and positional data
• 📷 Screenshot Only: Visual learning without position data
• 📍 Coordinates Only: Position-based learning without visual input

🏆 Reward Configuration

Configure positive or negative rewards for specific actions. Higher rewards encourage the behavior, while negative rewards discourage it.

⚡ Performance Considerations

• Higher screenshot resolutions increase VRAM usage and processing time
• GPU acceleration significantly improves training speed
• Monitor system resources during training

Technical Details 🔧

📚 Dependencies

• 🧠 DJL (Deep Java Library): Neural network framework
• 🔥 PyTorch: Backend neural network engine
• 🚀 CUDA Support: GPU acceleration for compatible hardware

💾 Data Storage

• Models are saved in .minecraft/mods/reinforcemc/
• Templates are stored in .minecraft/mods/reinforcemc/templates/
• Configuration files use JSON format

📊 Performance

• 💻 CPU Mode: Compatible with all systems, slower training
• 🖥️ GPU Mode: Requires CUDA-compatible GPU, faster training
• 🧠 Memory: Automatically manages memory allocation based on input configuration

Screenshots 📸

The mod includes a comprehensive GUI system with multiple configuration screens:

• 🏠 Main configuration menu
• 🏆 Reward system setup
• 📊 Input configuration
• 🎯 Training control panel
• 📈 Performance monitoring
• 📋 Template management

Compatibility 🔧

• 🎮 Minecraft Version: 1.16.5
• ⚡ Forge Version: 36.2.42+
• 💻 Operating System: Windows (primary), other platforms may work with CPU-only mode
• 💾 Memory: Minimum 4GB RAM recommended

License 📄

This project is licensed under the MIT License.

Note: This mod requires computational resources and is intended for users familiar with at least basic machine learning concepts. Training times can vary greatly depending on hardware configuration and complexity of the learning task.