-
Notifications
You must be signed in to change notification settings - Fork 0
strix omlx.stub
id: strix-omlx name: strix-omlx repo_path: ~/Documents/sw30labs/repos/strix-omlx remote_url: https://github.com/sw30labs/strix-omlx.git primary_language: Shell framework: Strix (LiteLLM) + OMLX category: ai-security stacks: [pentest, mlx, agentic] ingested: 2026-07-09 last_commit_date: 2026-07-05
A thin set of setup scripts that point Strix — the open-source autonomous AI security-testing agent — at a local OMLX MLX inference server, so it runs entirely on Apple Silicon against an abliterated ("Heretic'd") uncensored model. Keeps targets, prompts, and findings on the local machine; not a fork of Strix, just the wiring.
setup-strix-omlx.sh wires Strix to OMLX (local MLX, OpenAI-compatible) via LiteLLM openai/ routing at http://127.0.0.1:8000/v1; sibling scripts target Ollama and SGLang/NVIDIA DGX. run-strix.sh reads ~/.strix/cli-config.json. Model is MiniMax-M2.7-ultra-uncensored-heretic-oQ4-MLX, abliterated with Heretic to remove refusal directions so it cooperates with authorized offensive-security tasks. Strix decomposes an assessment into specialized agents (SQLi, IDOR, XSS, CSRF/business-logic, API) inside isolated Docker sandboxes.
Upstream usestrix/strix (Apache-2.0), LiteLLM, an OMLX server, Docker. Shell-based; no Python package of its own.
No CI workflows. Setup shell scripts only.
- strixresearch — companion benchmarks/comparisons of Strix across models and backends
- agentic-ai-pentesting — shared autonomous-pentesting research line
- local-mlx-responsesAPI-server — the kind of local OMLX backend this points Strix at
- Tags: [security, pentesting, mlx, apple-silicon, local-inference, llm]