AI Launchpad
Summon local models or launch online AI platforms. Local models require an application like Ollama or LM Studio to be running.
Local Models (ABL_SOVR)
Cloud Chat AIs (Online)
API & Agent Frameworks
Rig: Hardware Inventory
AI Server
GMKtec Evo X2 (Silent, always-on brain for LLMs).
Graphics Workstation
Intel NUC 12 Serpent Canyon (Powerhouse for Stable Diffusion & Skyrim).
Vision System
Ricoh Theta Z1 (360° eye for environmental analysis).
Audio Suite
Google Pixel Buds Pro (Discrete headset) & Beats Pill (room audio).
Input Devices
Folding Keyboard & Haute 42 Leverless Joypad (for macros/chording).
Power System
DJI Power 500/1000 & Zignes Solar Panel (for off-grid autonomy).
Connectivity
Portable Network Switch & optional Starlink (for "selectively connected" setup).
Toolkit: Software
AI "Players" / Interfaces
LM Studio
Power-user interface with granular hardware control. Your primary downloader.
Jan
Polished, user-friendly interface focused on the chat experience.
Ollama
Lightweight, "always-on" background engine for serving models via API.
Oobabooga's WebUI
Advanced "laboratory" for deep experimentation and fine-tuning.
Specialist Tools
Stable Diffusion WebUI
Your "Art Studio" for text-to-image generation. Runs on the Serpent Canyon.
Buzz
Your "Transcription Service" using the Whisper model for offline audio-to-text.
AnythingLLM / privateGPT
Future "Command Center" for RAG (chat with your documents).
Tasker (Android)
The automation tool for bridging your Ray-Ban glasses to your local AI rig.
Minds: Model Library
Heavyweight Generalists
WizardLM-2-8x22B (MoE)
Your crown jewel. Top-tier problem solver for complex reasoning.
DeepSeek R1 Distill Llama 70B
Heavyweight contender for deep reasoning and code analysis.
Llama 4:109B
Enthusiast-grade "heavy artillery" for maximum brain power.
Vision Specialists
Llava v1.6 34B
Your primary "AI with eyes" for deep, nuanced visual analysis.
MiniCPM-V
A smaller, highly efficient vision model for fast, "always-on" tasks.
Unfettered & Creative Specialists
WizardLM-Uncensored-StoryTelling-30B
Your dedicated creative writing expert for role-playing and narrative.
NeuralDaredevil-8B-Abliterated
Your benchmark for a truly "raw," unfiltered AI experience.
Dolphin Llama3 (Uncensored)
A legendary creative model, supercharged with multilingual capabilities.
Finance Hub
Trading Platforms & APIs
Robinhood
Your primary brokerage. Use the web app for desktop charting and trading.
Alpaca Markets
A modern platform designed for algorithmic trading with an excellent Python SDK.
Core Concepts
Momentum Strategy
A simple, effective strategy for the demo where the AI reacts to recent price trends.
The Two Dials
Control your system by adjusting **Frequency** and **Value** of trades in your scripts.
Paper Trading
The critical first step. Test your entire workflow with fake money before risking real capital.
Robotics Hub
Cloud & Local Frameworks
DeepMind Gemini Robotics
Google's advanced platform for experimenting with state-of-the-art cloud robotics.
ROS 2
The industry standard "operating system" for robotics that connects all your components.
NVIDIA Isaac Sim/Gym
Powerful simulation environments for testing robotics code in a virtual world.
Project Log: Conversation Archive
Initial Setup & Hardware
Discussions on PC choice, the "manufacturer first" driver installation method, and hardware troubleshooting.
Software Toolkit Deep Dive
Comparing LM Studio, Jan, and Ollama. Understanding their roles as "players" vs. "engines."
AI Model Curation
The process of finding, selecting, and understanding different AI models (Uncensored, Vision, Coding, etc.).
Advanced AI Concepts
Defining key terms like RAG, MoE, Quantization, and Knowledge Distillation.
Project Ideation
Brainstorming capstone projects like the "Workshop Guardian" and "AI Game Companion."
Networking & Synergy
The plan to connect the Evo X2 and Serpent Canyon to work together as a single, powerful rig.
Codex: Project Glossary
- ABL_SOVR
- The official name for your rig. Likely stands for Abliterated/Ablated, Sovereign. A streamlined, independent AI.
- DPO-Healed
- Direct Preference Optimization. A method of training an AI to better align with user intentions, making it more reliable. "Healed" implies it has been refined and improved through this process.
- GGUF
- The file format that allows large models to run efficiently on consumer hardware. The "fuel" for your engines.
- Quantization (e.g., Q4_K_M)
- Smart compression for AI models. Higher "Q" numbers mean higher quality but larger files. Q4_K_M is the universal "sweet spot."
- RAG (Retrieval-Augmented Generation)
- The process of letting an LLM access your personal documents to answer questions, making it an expert on your data.
- Uncensored / Abliterated
- Models fine-tuned by the community to remove safety filters and refusal behaviors, allowing for creative freedom.
- MoE (Mixture-of-Experts)
- A clever model architecture (like WizardLM-2-8x22B) that uses multiple smaller "expert" brains to achieve the power of a giant model with much better speed.
- Knowledge Distillation
- A training technique where a smaller "student" model learns the intuition and thought process of a larger "teacher" model, resulting in a highly efficient and capable AI.