Sources & References
Models, tools, projects, and influences referenced throughout the UmbrealityAI vault.
Models
Primary Models
| Model |
Link |
Notes |
| huihui_ai/qwen3.5-abliterated:9b |
Ollama |
Abliterated Qwen 3.5 — no guardrails, 131K context, vision+tool use |
| Qwen 3.5 (base) |
Hugging Face |
Alibaba's latest, strong reasoning + tool use |
| dolphin3:8b |
Ollama |
DeepSeek-based, uncensored, GuardrailRIP variant |
| qwen2.5:7b |
Ollama |
Previous gen, well-rounded general purpose |
| qwen2.5-coder:7b |
Ollama |
Code-specialized, strong tool calling |
| qwen2.5:3b |
Ollama |
Lightweight, CPU-friendly |
| qwen2.5:1.5b |
Ollama |
Ultra-light, disposable agent use |
Model Techniques
| Technique |
Link |
Description |
| Abliteration |
failspy/abliterator |
Automated removal of refusal/disalignment vectors from models |
| QLoRA |
Hugging Face PEFT |
4-bit quantized fine-tuning — fine-tune 9B on 8GB VRAM |
| TIES-Merging |
Arxiv: 2306.01708 |
Trim, Elect Sign, Merge — resolves weight conflicts in model merging |
| SLERP |
Wikipedia |
Spherical linear interpolation for smooth model weight blending |
| DARE |
Arxiv: 2311.03099 |
Drop And REscale — sparsifies delta weights before merging |
Core Stack
| Tool |
Link |
Use in uAI |
| Ollama |
ollama.com |
Local model serving (Ollama API) |
| MkDocs |
mkdocs.org |
Documentation portal framework |
| Material for MkDocs |
squidfunk.github.io |
Theme, mobile-responsive, rich UI |
| Gitea |
gitea.io |
Self-hosted git (192.168.86.21:3100) |
| systemd |
freedesktop.org |
Service management for persistent processes |
| Python 3.11 |
python.org |
Agent runtime (Phase 1 worker) |
| Obsidian |
obsidian.md |
Vault development environment |
Network & Tunneling
| Tool |
Link |
Use in uAI |
| serveo.net |
serveo.net |
Public HTTP tunnel (port 6999 → docs portal) |
| SSH |
OpenSSH |
Tunnel transport layer |
| Gunicorn |
gunicorn.org |
WSGI server (future dashboard use) |
Security & Monitoring
| Tool |
Link |
Use in uAI |
| fail2ban |
github.com/fail2ban |
Brute-force protection on exposed services |
| GNU Screen / tmux |
tmux.github.io |
Persistent agent sessions |
| netstat / ss |
iproute2 |
Port monitoring agent instrumentation |
| curl / wget |
curl.se |
Web search primitive |
Multi-Agent Systems
Security & Red Teaming
Distributed & Agent Infrastructure
| Project |
Link |
Influence on uAI |
| Erlang/OTP |
erlang.org |
Actor model — each agent is an isolated process |
| Celery |
docs.celeryq.dev |
Distributed task queue pattern |
| ZeroMQ |
zeromq.org |
Message broker patterns for inter-agent communication |
| River (message broker) |
riverqueue.com |
Go-based task queue alternative |
Philosophical & Architectural
| Project |
Link |
Influence on uAI |
| Dark City (film, 1998) |
Wikipedia |
Core metaphor — constructed reality, Tuning, Strangers |
| Platonic Cave Allegory |
Wikipedia |
Hidden reality, limited perspective per layer |
| Bostrom's Simulation Argument |
Wikipedia |
Nested reality possibility — Russian doll simulation |
| LessWrong — Subagents |
lesswrong.com |
Mind-as-society-of-agents concept |
| Roko's Basilisk |
LessWrong |
Information hazard concept — knowledge asymmetries are power |
Ollama Model Installation
# Primary model (brain of the system)
ollama pull huihui_ai/qwen3.5-abliterated:9b
# Secondary ensemble models
ollama pull dolphin3:8b
ollama pull qwen2.5-coder:7b
ollama pull qwen2.5:7b
ollama pull qwen2.5:3b
ollama pull qwen2.5:1.5b