Project currently on hold while I am dealing with medical issues.
I will most likely convert this project into a local home assistant ran on a single board computer like the Orange Pi 5.

Aetherius – A Personal Digital Assistant

Aetherius is a versatile, modular AI Assistant/Sub-Agent Framework that adapts to your needs. Its capabilities extend beyond conventional chatbots:

Real Time Data: Aetherius has access to search the web or your own data in agent mode, allowing for information that isn’t contained in the base model.

Multi-Agent Framework: Aetherius gives you the ability to create sub-agents for whatever use case you have. Alternativly you can have it trigger python scripts instead.

Reflective Journal: Speak your mind freely and receive thoughtful feedback without judgment or fear.

Learning Tool: Dive deep into your favorite topics and enhance your knowledge effortlessly.

Data Analysis Companion: Harness the power of your data with Aetherius by your side.

Cognitive Offload: A second brain that’s entirely private, aiding you in organizing thoughts and ideas.

Content Generation: Easily Generate Content based off of files or webscrapes.

Current Tools

With Aetherius, you have an arsenal of tools to explore and use:

• Web Scrape/Search: Gather information from websites with ease.
• File Processor: Process a variety of file types for insights and knowledge. The supported file types are: .epub, .pdf, .txt, .png, .jpg, .jpeg, .mp4, .mkv, .flv, and .av
• Photo Analysis: Aetherius can see photos you send and complete tasks based on them.

Customize Your Experience

Main Aetherius Chatbot: A framework for the creation of custom sub-agents for Aetherius.

• Forced Memory Mode: Aetherius will always upload it’s memories.
• Auto Memory Mode: Aetherius autonomously manages memory uploads.
• Manual Memory Mode: You decide when to upload memories.
• Training Memory Mode: Control memory uploads for each memory type.
• Agent Mode: Activate the use of any Sub-Agents for use in Aetherius’s agent loop

Current Sub-Agents

• External Resources: Will search Aetherius’s External Resource Database. (Database from Web and File Scrapes) If the information cannot be found, it will do a simple websearch for the information. You can disable the websearch and change the engine in the script file.
• Implicit Memory Search: Will search Aetherius’s Implicit Memories to complete the task.
• Explicit Memory Search: Will search Aetherius’s Explicit Memories to complete the task.
• Episodic Memory Search: Will search Aetherius’s Episodic Memories to complete the task.
• Flashbulb Memory Search: Will search Aetherius’s Flashbulb Memories to complete the task.

Core Features:

1. AI-Driven Gameplay:
   • The engine harnesses the power of LLMs to create dynamic and engaging game narratives.
   • Player choices and interactions are at the forefront, with the AI adapting and evolving the story and world in real-time.
2. Updateable Environment:
   • A key feature of this engine is its ability to modify game maps and settings based on player actions and decisions.
   • This dynamic environment ensures a unique and personalized gaming experience, with every decision impacting the game world.
3. Realistic NPC Interactions:
   • All non-player characters (NPCs) within the game will be equipped with a simplified version of my Aetherius system.
   • This will allow NPCs to engage in realistic conversations, react authentically to player actions, and contribute meaningfully to the game’s narrative.

Goal: The primary aim of this project is to create a JRPG engine that is not only robust and versatile but also intuitive and user-friendly. By integrating AI into the core gameplay, the engine hopes to deliver a text RPG experience that is both immersive and endlessly captivating.

[Current Progress/RoadMap]

Unfortunately I have lost all progress due to my SSD failing. For now I will be focusing on remaking all of the core systems. All buttons, menus, and images are placeholders for now. Once all core systems are in I will make it look nice.

Main Menu – Done

Map Loading and Player Movement – Done

Visualized Map to move on – Done

Map Tile Resizing when screen size is changed – Done

Player Sprite and Movement – Done

Player Loading/Saving – Done

Tile Description Window – Done

Tile Linking for Switching Maps – Done

Player Information Page: Done

Player Skill Page: Done

Player Inventory System and Page: Done

Dialogue System: Done

Battle System: Done

Linking Script activation to skill usage in modular system: In Progress

Non-Combat Skill Usage: Not Started

Getting skills to edit the terrain: Not Started

Editing Terrain Based on Player Actions: Not Started

Each book featured here is part of my personal collection, all of which were published before 1924 and have since entered the Public Domain.

List of Books currently awaiting Digitization:

Myths of the Hindus and Buddhists by Nivedita and Coomarasawamy

Great Leaders and National Issues of 1896 by the Non-Partisan Bureau of Political Information

The Lost Dryad by Frank R. Stockton

The Rosicrucians, Their Rites and Mysteries-Fifth Edition by Hargrave Jennings

🔐 True End-to-End Encryption (E2EE) & Integrity

Every API call and WebSocket message can be encrypted using a session-specific key established via an ECDH key exchange. The server is authenticated using ECDSA signatures to prevent Man-in-the-Middle (MitM) attacks, and session keys are periodically ratcheted for post-compromise security. Subresource Integrity (SRI) hash verification ensures that the frontend code matches expected build hashes before establishing communication.

🗄️ Zero-Knowledge Encrypted VFS & MFA

A client-side encryption layer secures your files. A session master key is derived from your password using a custom, dependency-free JavaScript implementation of Argon2id utilizing 64-bit BigInt operations (replacing old PBKDF2 schemes). Files are encrypted in your browser before they are sent to the server. Key retrieval and unlock sessions are optionally guarded by Multi-Factor Authentication (MFA/TOTP).

🛡️ Multi-Layer Sandboxing

• Frontend Web Worker Isolation: Applications run inside dedicated Web Worker threads, completely isolated from the DOM and global window object.
• Canvas Compositor: UI is driven via a canvas-native widget library, which serializes drawing instructions to a Canvas Compositor thread that renders widgets and forwards user inputs back to the worker.
• Rust WASM Microkernel: Offloads path-traversal checking, memory boundaries, and VFS capability checking to a client-side WebAssembly binary compiled from Rust.
• WebAssembly Bytecode Instrumentation: User WASM binaries are rewritten at load-time to inject env.checkpoint hooks at block entries and loop backedges, allowing cooperative yielding and execution-limit enforcement via WasiBridge.
• Backend Isolation: Custom Python services are packaged, built into individual Docker container images (user-app-<app_key>:latest), and executed in ephemeral, network-isolated environments with strict resource constraints.

🧩 Extensible App Platform

Install custom apps bundled as .zip archives containing manifests, workers, and Python backend scripts. App capabilities, resource permissions (e.g. VFS scope, authorized syscalls), and custom Python endpoints are defined declaratively in manifest.json.

🧠 Built-in AI & RAG Engine

Web-OS features native server-side AI integration, orchestrating a Qdrant vector database, sentence-transformers, and BM25 rank scoring for Retrieval-Augmented Generation (RAG). Applications can make LLM_GENERATE system calls to interface with local (Koboldcpp) or external (OpenAI, Anthropic) models.

---

A self-contained edge-computing field server

⛺ Off-Grid Amenities
Local Intranet: A “Captive Portal” style dashboard that greets users when they connect.
LAN Chat: A WhatsApp-like chat server that works without internet (great for communicating between tents/vehicles).
Media & File Server: Local NAS for hosting movies, maps, and sharing files between devices.

🛡️ Security & Surveillance
AI Object Detection: Uses the NPU (Neural Processing Unit) to detect humans or animals in real-time.
Local NVR: Ingests RTMP feeds from Wi-Fi cameras (GoPro, Phones, IP Cams).
Threat Dashboard: A web-based UI accessible to anyone on the Wi-Fi to see live feeds and security alerts.

🛠️ Current Hardware Stack
SBC: Orange Pi 5 (RK3588)
OS: Ubuntu 22.04 (Joshua-Riek Rockchip build).
Network: USB-to-Ethernet Bridge + TP-Link TL-WR902AC (AP Mode).