How vedalai/neuro-amongus works
Neuro is an Among Us plugin that exchanges game-state observations and neural-network actions with an external AI, while also recording gameplay for model training. In play, the plugin sends frames over a WebSocket and applies received movement/action commands; local handlers support navigation, interactions, minigames, and impostor behavior. Recording assembles game-state domains into frames for upload and offline conversion/training. The graph retains documented AI and gameplay capabilities, but several subsystem links are omitted where source was not sampled and the index does not establish their runtime calls.
Main entry points: CommunicationHandler.cs AI communication · Frame.cs Frame assembly · MovementHandler.cs Movement control · MinigameHandler.cs Minigame handler
Components 25
Plugin runtime
- Plugin bootstrap
Neuro/NeuroPlugin.cs - AI communication
Neuro/Communication/AmongUsAI/CommunicationHandler.cs - WebSocket worker
Neuro/Communication/AmongUsAI/WebSocketThread.cs - Event dispatch
Neuro/Events/EventManager.cs
Gameplay control
- Movement control
Neuro/Movement/MovementHandler.cs - Pathfinding
Neuro/Pathfinding/PathfindingHandler.cs - Console discovery
Neuro/Movement/ConsoleFinder.cs - Game interactions
Neuro/Interactions/InteractionsHandler.cs - Minigame handler
Neuro/Minigames/MinigameHandler.cs - Minigame solvers
Neuro/Minigames/Solvers - In-game cursor
Neuro/Cursor/InGameCursor.cs - Impostor behavior
Neuro/Impostor/ImpostorHandler.cs
Recording
- Frame assembly
Neuro/Recording/Frame.cs - Player recording
Neuro/Recording/LocalPlayer/LocalPlayerRecorder.cs - Task recording
Neuro/Recording/Tasks/TasksRecorder.cs - Game-state recorders
Neuro/Recording - Recording uploader
Neuro/Recording/Uploader.cs
AI training
- Recording conversion
AI/data/converter.py - Game data model
AI/data/game.py - Training dataset
AI/nn/dataset.py - Neural network
AI/nn/model.py - Model training
AI/train_model.py - Model fine-tuning
AI/finetune_model.py
Other components
- Among Us player
- External AI service
Connections 23
- Plugin bootstrap → AI communication: initializes
Neuro/NeuroPlugin.cs - AI communication → WebSocket worker: starts and uses
Neuro/Communication/AmongUsAI/CommunicationHandler.cs - AI communication → External AI service: exchanges frames
Neuro/Communication/AmongUsAI/CommunicationHandler.cs - AI communication → Movement control: sets move direction
Neuro/Communication/AmongUsAI/CommunicationHandler.cs - Movement control → Pathfinding: requests paths
Neuro/Movement/MovementHandler.cs - Movement control → Console discovery: finds consoles
Neuro/Movement/MovementHandler.cs - Game interactions → Minigame handler: checks and launches
Neuro/Interactions/InteractionsHandler.cs - Minigame handler → In-game cursor: controls
Neuro/Minigames/MinigameHandler.cs - Minigame handler → Minigame solvers: dispatches solver
Neuro/Minigames/MinigameHandler.cs - Minigame solvers → In-game cursor: performs inputs
Neuro/Minigames/Solvers/FuelEnginesSolver.cs - Impostor behavior → Movement control: waits and moves
Neuro/Impostor/ImpostorHandler.cs - AI communication → Event dispatch: registers handler
Neuro/Communication/AmongUsAI/CommunicationHandler.cs - Frame assembly → Player recording: collects frame data
Neuro/Recording/Frame.cs - Frame assembly → Task recording: collects frame data
Neuro/Recording/Frame.cs - Frame assembly → Game-state recorders: collects game state
Neuro/Recording/Frame.cs - AI communication → Frame assembly: serializes current frame
Neuro/Communication/AmongUsAI/CommunicationHandler.cs - Task recording → Game data model: supplies task observations
- Recording uploader → Recording conversion: provides recordings
- Recording conversion → Game data model: converts data
- Game data model → Training dataset: builds examples
AI/nn/dataset.py - Training dataset → Model training: feeds training
- Model training → Neural network: trains
- Neural network → Model fine-tuning: fine-tunes
AI/finetune_model.py