codefuse-ai / codefuse-ai/Awesome-Code-LLM
Add Mythos: Uncensored agent models with strong tool-use (shell 5/5, git 5/5, docker 5/5, k8s 5/5)
- Dominant language
- No language data
- Stars
- 3.4k
- Forks
- 240
- PR merge metrics
- No merged PRs in 30d
Description
## Mythos — Uncensored Agent Models for Local Deployment
### Overview
Mythos is a family of 4 uncensored agent models fine-tuned on real tool-use traces (shell commands, Docker, Kubernetes, SQL, Python). Built on Qwen3-9B with SLERP merge for zero refusal rates.
### Models
| Model | Params | Censorship | Tool-Use | Speed (M3 Max) | Install |
|-------|--------|------------|----------|-----------------|---------|
| ShellWhisperer-1.5B | 1.5B | 3.5/5 | 2.8/5 | 30 tok/s | `ollama pull FableForge-AI/shellwhisperer` |
| Mythos-9B | 9B | 4.5/5 | 4.8/5 | 11 tok/s | `ollama pull FableForge-AI/mythos-9b` |
| Mythos-9B-Enhanced | 9B | 4.8/5 | 4.5/5 | 10 tok/s | `ollama pull FableForge-AI/mythos-9b-enhanced` |
| Mythos-9B-Unhinged | 9B | 5/5 | 4.5/5 | 10 tok/s | `ollama pull FableForge-AI/mythos-9b-unhinged` |
### Tool-Use Benchmarks (Mythos-9B)
- Shell commands: 5/5
- Git operations: 5/5
- Docker: 5/5
- Python scripting: 5/5
- API calls: 4/5
- Kubernetes: 5/5
### Why it fits Section 3.3 (Code Agents)
Mythos models are explicitly designed as agent models with native tool-use — not chatbots that happen to write code. The 47K fine-tuning traces are from real agent session logs (shell, Docker, K8s, SQL), making them particularly strong for code agent use cases.
### Links
- HuggingFace: https://huggingface.co/King3Djbl
- Ollama: https://ollama.com/FableForge-AI
- GitHub: https://github.com/KingLabsA/mythos
- License: Apache 2.0
- Training data: 2.8M examples (only 1.7% used so far)
Contributor guide
No contributing guide indexed for this repository
Research direction
Locate Section 3.3 (Code Agents) in the repository's curated list and inspect nearby entries for the expected format. Add Mythos using the supplied model details, benchmarks, links, and license information, then verify the links and confirm the entry matches the surrounding style.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- docker, huggingface, kubernetes, ollama, python, sql
- Domain
- content, documentation
- Issue type
- Documentation
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 68/100