Detect OMLX as an MLX backend and fix MLX filter results

[bruvv]

When running a mac, https://omlx.ai/ is pretty good solution. Friendly admin ui and inbuild downloader that supports MLX:

The thing is though llmfit does not show me MLX things it seems. I have set the following:

i pick the first one:

That gives me the following download: https://huggingface.co/mradermacher/Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-Distilled-GGUF (will be great if this is clickable btw)

But when searching this supposed MLX format i get:

When I uncheck MLX i do see it:

What is going on?

Ps please add omlx also as MLX detector in the top status bar of llmfit:

[AlexsJones]

Thanks @bruvv — splitting this into the two things it contains:

1. OMLX detection (feature request): treating OMLX as an installed MLX-capable runtime. Reasonable ask for Apple Silicon users.
2. MLX filter showing no results (possible bug): could you share your llmfit version (llmfit --version) and what the hardware summary reports for your Mac (chip / unified memory)? If MLX models are missing entirely with the MLX filter on, that may be the conservative KV-cache filtering tracked in MacBook Neo shows no models under "Good" or "Runnable" due to conservative KV cache filtering #635 or a catalog gap rather than the filter itself.

Treating the OMLX integration as the feature here; the filter half is needs-info.

🤝 Handled by Alex's Repo Steward — replies reviewed by @AlexsJones.

Learn more or run your own: https://github.com/AlexsJones/repo-steward