whereismyai
MLX Direct Decider
MLX Direct Decider is an open-source Python library for training-free decisions from existing MLX models on Apple Silicon.
MLX Direct Decider is an open-source Python library for training-free decisions from existing MLX models on Apple Silicon Macs. It enables local action selection, classification, semantic routing, and enum or boolean decisions using existing checkpoints. The library reads candidate logits directly from the unchanged output head of an MLX-LM or MLX-VLM model without generating a continuation. Applications supply state, a natural-language request, and allowed choices to receive a stable candidate ID, nomatch, or abstain. It integrates through a Python API, a command-line interface, or a local text HTTP service.
MLX Direct Decider at a glance
Key features
- Selects actions without fine-tuning or extra weights
- Supports MLX-LM checkpoints for text decisions
- Supports MLX-VLM checkpoints for text and image decisions
- Returns explicit nomatch or abstain outcomes
- Provides Python API, CLI, and localhost HTTP service
- Reuses stable context prefix while evaluating requests
- Ranks choices using candidate scores from logits
Use it for
- Select items or open features in local applications
- Perform semantic routing and classification
- Answer boolean and enum questions
- Control local games using structured board data
Alternatives to MLX Direct Decider
- Laya Mlx — Provides local MLX inference with no PyTorch or cloud API dependency.
- Rapid-MLX — Run 250 MLX-native models locally on Apple Silicon with high inference speeds.
- Decisions API — Fast classification and routing decisions without generating free text.
- MTPLX — It runs local models twice as fast via native MTP speculative decoding.
- oMLX — It uses paged SSD KV caching to reduce agent TTFT to under 5 seconds.
- MLflow — Complete observability and evaluation for production AI apps.
- Nouli — It gives stateless decision models a persistent memory across branching sessions.
- Ollaya — Run decision models locally with typed, calibrated answers in milliseconds.
- MLX LM — Runs LLMs and fine-tuning natively on Apple silicon hardware.
- MLC LLM — Enables native optimization and deployment of AI models across platforms.
- llm-d — Deploys production-grade LLM serving on existing hardware clusters.
- LiteLLM — One OpenAI-format API for 100+ LLM providers
All MLX Direct Decider alternatives