Show HN: MLX Direct Decider, Mac local decider without fine-tuning

Category: ai-ml

Tags: mlx, apple-silicon, local-llm, decision-engine, python, semantic-routing, training-free

Score: 6.0/10 (Innovation: 6, Technical: 6, Documentation: 7, Utility: 5)

MLX Direct Decider is a Python library that turns existing MLX-LM/MLX-VLM checkpoints into a local decision engine on Apple Silicon by scoring candidate logits from the model's original output head, without fine-tuning. It supports action selection, semantic routing, enum/boolean decisions, and explicit no-match/abstain outcomes via Python API, CLI, and localhost HTTP. Notable for a pragmatic training-free approach to local LLM-driven decisioning with honest scoped benchmarks.

Target audience: backend devs, ML engineers, Apple Silicon developers building local AI agents or game/UI automation

Repository: https://github.com/ItsOdeLeo/MLX-Direct-Decider · Python · MIT · 5 stars

View on Hacker News