
projectBEA
BeaCraft is a Fabric mod bridge. It exposes game state and player actions via WebSocket, enabling autonomous LLM agents to play Minecraft.
- Загрузки
- 336
- Подписчики
- 0
- Обновлён
- 18 февраля 2026 г.
- Лицензия
- MIT
Опубликован 18 февраля 2026 г.
BeaCraft: Technical Bridge for Autonomous Agents
BeaCraft is a Fabric mod that functions as a WebSocket interface between Minecraft and external AI systems. It provides the necessary infrastructure for LLM-based agents to perceive the game world and execute complex tasks programmatically.
Core Functionalities
- State Broadcasting: Periodically transmits JSON packets containing player health, hunger, coordinates, inventory, and surrounding block/entity data.
- Action Execution: Maps external command packets to in-game behaviors including pathfinding, mining, crafting, and combat.
- Asynchronous Feedback: Returns status updates (IDLE, FINISHED, INTERRUPTED) to the agent, allowing for reactive logic and error handling.
- Vision & Communication: Supports remote screenshot requests for visual context processing and bridges the in-game chat for social interaction.
Technical Architecture
- Protocol: WebSocket (default
ws://localhost:8080). - Format: Standardized JSON envelopes for both state updates and action requests.
- Compatibility: Specifically designed for integration with Python-based AI frameworks like ProjectBEA.
Implementation Requirements
- Loader: Fabric Loader.
- Configuration: Managed via
config.jsonto define server ports and update frequencies. - Environment: Operates as a local bridge, requiring the external agent to run on a reachable network address.
Ченджлог
1.0.0Бета1.21.9, 1.21.10, 1.21.11 · 18 февраля 2026 г.
Changelog - BeaCraft 1.0.0
Initial Release
Core Infrastructure
- Established WebSocket server (default port 8080) for external agent synchronization.
- Implemented bidirectional JSON communication protocol for state and commands.
- Added background thread management for non-blocking game-to-agent data flow.
State Perception
- Implemented full player state broadcasting (Health, Hunger, Position, Rotation).
- Added Inventory monitoring and Equipment tracking.
- Integrated Surroundings scanner for real-time block and entity detection.
Action System
- Developed core toolset for LLM agents: mine_block, move_to, attack_entity, and craft_item.
- Implemented pathfinding logic for autonomous navigation.
- Added event feedback loop (SUCCESS, FAILURE, INTERRUPTED) for action validation.
Vision and Chat
- Added remote screenshot request capability for vision-based processing.
- Integrated chat bridge for reading and sending messages via WebSocket.
Configuration
- Added
config.jsonfor custom server URL and port definitions. - Implemented debug logging for troubleshooting agent-mod communication.
Комментарии
Загружаем…