Rapid-MLX Release Notes

raullenchai · Run local AI models, chat, and manage a model server on an Apple Silicon Mac.

Website downloadAI ToolsFree and open sourceApple silicon · Intel

Download Homepage

Downloads come from github.com · OpenNavo hosts no installers

1.0.0
Latest
OpenNavo editorial
Adds agent chats and new models, with simpler compute sharing and safer public routes.
Adds persistent terminal agent chats with tools, approvals and images; Desktop harness chats are experimental and off by default. Terminal commands are not sandboxed.

Adds LFM3, LFM3-VL, Ornith and MiniMax-H3 support. MiniMax-H3 needs 160 GB+ unified memory and BF16 weights.

Public compute shares no longer expose local control routes. Desktop sharing uses one QuickSilver key and a setup checklist.

Chat messages support up to 8 images, with optional release of idle models. Improves reasoning and tool-call parsing.

Official release notes
0.16.10
Patch
OpenNavo editorial
CompatibilitymacOS 14 or later · Native on Apple silicon
Rapid-MLX improves dictation and model downloads and adds three model choices.
Starts dictation capture immediately, avoiding lost audio at the beginning of a sentence.

Lets covered models download through the mirror without first contacting Hugging Face, and uses complete cached transcription models and processor assets when that network is unavailable.

Adds Nanbeige4.3-33B, Spark-X2.6-4B, and Sage-A1.5. Speeds up LFM2 MoE generation on the tested Apple Silicon systems.

Official release notes