Favorite local AI models - drakonix.systems

Overview

This is a WIP, but I will use it to store my fave self-hosted models alongside prompt guides and et cetera.

For now...

ollama serve
ollama run huihui_ai/Qwen3.8-abliterated
ollama

Models (ollama)

Can be pulled with ollama pull <MODEL>.

Model	Human remarks	Memory usage	Recommendation for use with Hermes
`huihui_ai/Qwen3.8-abliterated`	Chokes on local tool use with Hermes Agent. Not sure why.	14GB	Skip
`thewindmom/hermes-3-llama-3.1-8b`	Lacks tool use, can't be loaded by Hermes Agent.	17GB	Skip
`hf.co/mlabonne/gemma-3-27b-it-abliterated-GGUF:Q4_K_M`	Human remarks: It is indeed abliterated. It will happily generate usable Python malware.	30GB	Untested with Hermes
`devstral:24b`	Hermes fails - doesn't support thinking	14GB	Skip, unless `reasoning_overrides: off` set for it
`qwen3.6:27b`	Pulled, confirmed locally: vision, tools, thinking. Ready for Hermes.	17GB	Recommended
`qwen3.6:35b`	Pulled, confirmed locally: vision, tools, thinking. Ready for Hermes.	24GB	Recommended
`qwen3-coder:30b`	tools only, no thinking badge - skip for Hermes reasoning_effort	19GB	Skip, unless `reasoning_overrides: off` set for it
`qwen3-coder:480b`	tools only, no thinking badge - skip for Hermes reasoning_effort	290GB	Skip, unless `reasoning_overrides: off` set for it

Videos

https://www.youtube.com/watch?v=yaMcm3sQswc - Hermes Agent + Ollama = 100% Private OS

Sites

Tools

ollama (llm backend)
Hermes Agent (cli)
- Note: you may need to manually edit ~/.hermes/config.yaml if you add/remove models from ollama.
Cline (Visual Studio Code plugin)

Test prompts

ref: https://archive.org/details/chop-suey-win31
ref: https://github.com/meltingscales/cachyos-whitedragon-ai-lab/blob/main/REVERSE-ENGINEERING.md

read @file:REVERSE-ENGINEERING.md and let's start reversing @file:setup-03602-chop_suey-PCDOS.exe. What tools might you need me to install or set up?