Awesome MLX — curated list of 80+ MLX projects, tools, and resources

[raullenchai]

I've been maintaining a curated list of the MLX ecosystem: Awesome MLX

It covers 80+ projects organized into categories:

• Core Framework — mlx, mlx-lm, mlx-swift, node-mlx
• Inference & Serving (12 projects) — with a "which one should I use?" quick guide
• Training & Fine-tuning (7) — mlx-tune, MLX-GRPO, TransformerLab, etc.
• Audio & Speech (12) — mlx-audio, lightning-whisper-mlx, voicebox, etc.
• Image & Video — mflux, mlx-video, flux.swift
• Vision & Multimodal — mlx-vlm, ml-aim
• Embeddings & RAG — mlx-embeddings, VecturaKit, jina-grep-cli
• Swift Ecosystem (7 native apps) — Klee, osaurus, fullmoon-ios, etc.
• Models — with a RAM-based recommendation table (8GB/16GB/32GB/64GB+)

I also added a Quick Start section ("I just want to chat" / "I need an API server" / "I want to fine-tune" / "I'm building a Swift app") and a low-friction submission flow — anyone can add their project via a simple issue form.

Would love to get this in front of more MLX developers so it becomes a go-to resource. PRs and suggestions welcome!

https://github.com/raullenchai/awesome-mlx

[ynikitenko]

Hi @raullenchai , thanks for your great work!

I'm considering using MLX and would like to ask you: Do there exist simple data analysis tools for MLX? Like histograms, plots, averages, etc. mlx-data looks rather general (not only for ML), so I'm surprised your list does not include any.

UPD: I didn't notice it was based on this list, #654 .

[rajveer43]

This is a great resource — thanks for building it. One gap I noticed: the list has a couple of KV-cache-related entries (TurboQuant-MLX, SwiftLM) but no dedicated compression library. I maintain VeloxQuant-MLX — 43 KV-cache compression methods for mlx_lm (quantizers, token-eviction, cross-layer merging), each adapted from a published paper, with hand-written Metal kernels for the hot path. Filed a submission via your issue form: raullenchai/awesome-mlx#29.