How libraryofcelsus/aetherius_ai_assistant works
Project’s Nature and Goals
• Project Type: A modular, locally operated AI assistant and multi-agent framework.
• Purpose: To provide a digital personal assistant with advanced memory management (explicit, episodic, etc.), multi-agent processing, external data integration (web scrapes, file processing, OCR), and support for multiple AI backends (Oobabooga, KoboldCpp, OpenAI, AetherNode). The system is designed to be both a chatbot and an orchestrator that handles complex agent loops and memory operations.
File Structure for Key Components
• Core Engine – Aetherius_API:
• Contains the main entry script (Main.py), API settings, configuration files (chatbot_settings.json), languages for generation (Generation_Settings), and model format definitions.
• Subdirectories like “Sub_Agents” (for various agent implementations: Llama_2, KoboldCpp, Oobabooga, OpenAi) encapsulate different agent modules.
• “Cloning”, “Examples”, “Tools”, “Utilities”, and “resources” indicate auxiliary functions that support data conversion (e.g., audio conversion), web scraping, memory search strategies, and integration with external libraries.
• User Interface – Aetherius_GUI and Terminal/CLI Scripts:
• Provided by the Aetherius_GUI folder and several run_*.bat / shell scripts for launching different modes, including a dedicated terminal UI.
• External Integrations and Configuration:
• API keys folder and configuration scripts (like API_Settings.json) reveal integration with external services (e.g., OpenAI, Qdrant, Tesseract, Ngrok).
• Database and Memory Storage:
• Qdrant integration and file uploads (Upload folder) suggest a dedicated layer for long-term memory storage and retrieval.
Key High-Level Components
A. Frontend Layer:
– User Interfaces (GUI, Terminal, Discord Bot, possibly Gradio UI).
– Input layer capturing user requests.
B. Main AI Engine (Aetherius_API):
– Input Processing Module: processes user input, handles input expansion and conversation history.
– Core Agent Architecture: the flow with multiple stages:
• Knowledge Domain Extraction
• Semantic Term Separation
• First Memory DB Search (for inner monologue generation)
• Inner Monologue Generation and Second Memory DB Search (for intuition generation)
• Implicit Memory Generation and Final Response Generation
• Explicit and Episodic Memory Generation and subsequent updates.
– Sub-Agent Manager: Handles the dynamic invocation of sub-agents for specialized tasks (e.g., External Resource Search, Memory Search types like episodic or flashbulb).
– Memory and Task Management: Modules for memory consolidation (implicit, explicit, episodic) and tasklist generation.
C. Tools and Auxiliary Services:
– Web Scrapers, File Processors, and OCR modules (within the Tools directory and Utilities).
– External Script Triggers: Additional isolated scripts from Cloning or Examples folders that complement processing (e.g., converting audio files).
D. Database and External APIs:
– External AI Model Providers: Represent integrations with OpenAI, KoboldCpp, Oobabooga, and AetherNode for LLM inference.
– API Gateway (API_Settings.json and Discord Bot trigger): Showing how the service can expose itself externally when needed (e.g., via Ngrok or a dedicated REST API).
E. Configuration and Deployment:
– Configuration Files (chatbot_settings.json, various settings.json files under Generation_Settings and model formats).
– Installer Scripts and Docker for Qdrant, showing containerization/deployment aspects.
Representing Relationships and Data Flow
• User Input (via UI) → AI Engine → Memory Modules → Sub-Agent invocations → External Data/Tools → Final Response.
– Break out the agent loop components (e.g., “Input Expansion” to “Inner Monologue Generation” to “Tasklist Generation”) as sequential modules.
– Indicate interactions between modules (e.g., the Sub-Agent Manager drives calls to different Memory Search scripts in the “Sub_Agents” folder).