MLX not detected?

[imavroukakis]

Hey folks! I've installed mlx and mlx-lm both via brew and pipx , but llmfit is not detecting the framework in either scenario - what am I doing wrong?

Thanks!

[AlexsJones]

Hi @imavroukakis — you're not doing anything wrong; it's how the detection probe works that's biting you.

On macOS, llmfit marks MLX as available if either of these is true (llmfit-core/src/providers.rs):

1. A running MLX server responds — it probes MLX_LM_HOST if set, then http://localhost:8080 (the mlx_lm.server default), then http://127.0.0.1:8000 for oMLX identity.
2. python3 -c "import mlx_lm" succeeds — i.e. the mlx_lm package is importable by whatever python3 is first on your PATH.

Both of your install routes fail check 2 by design:

• pipx installs mlx-lm into its own isolated venv — that's the whole point of pipx — so the system python3 can't import mlx_lm, even though the mlx_lm.server / mlx_lm.generate commands are on your PATH.
• brew links the package against Homebrew's own Python; if python3 on your PATH resolves elsewhere (e.g. the Xcode Command Line Tools python), the import fails there too. You can check with python3 -c "import mlx_lm" — if that errors in your terminal, llmfit sees the same thing.

Two easy ways to get detected:

• Start the server (works fine with your existing pipx install): mlx_lm.server — llmfit will find it on localhost:8080, or set MLX_LM_HOST=http://localhost:PORT for a custom port. This is also the more functional path, since llmfit can then list the models the server actually serves.
• Or make the import work: python3 -m pip install mlx-lm into the interpreter that's first on your PATH (or activate a venv that has it before running llmfit).

One more note: your downloaded MLX models (anything in the HuggingFace cache from mlx-community or *-mlx repos) are picked up by a separate cache scan, so those can show up even before the runtime itself is detected.

If python3 -c "import mlx_lm" succeeds and llmfit still doesn't show MLX, that would be a real bug — please say so here and I'll dig in.

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

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

[imavroukakis]

Cheers @AlexsJones Repo Steward 😃 - the mlx_lm.server direction worked. I'm curious as to why the homebrew install did not though, given that

which python3
/opt/homebrew/bin/python3

/opt/homebrew/bin/python3 -c "import mlx_lm"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
    import mlx_lm
ModuleNotFoundError: No module named 'mlx_lm'

🤷

Problem for another day I suppose - because now, the models themselves cannot be found