local AI runtime

A local AI runtime for Apple Silicon.

Run models, fine-tune LoRA, benchmark, and evaluate—on one local Mac workflow.

Get started on GitHub Read setup guide View current status

No external frontend runtime libraries are used.

Local AI is powerful. Operating it is fragmented.

Why this exists

Model serving, tuning, benchmarking, and evaluation are usually split across terminals, scripts, and notebooks. Melix compresses them into one local-first runtime.

Product Loop

Serve. Fine-tune. Benchmark. Evaluate.

Serve

Run local model sessions on Apple Silicon.

Fine-tune

Train and manage LoRA adapters locally.

Benchmark

Measure model and adapter behavior with repeatable runs.

Evaluate

Compare outputs, results, and workflow quality over time.

A runtime layer between models and workflows.

Serve → Fine-tune → Benchmark → Evaluate

Model Providers

ggml, GGUF, LoRA adapters

Melix Runtime

Session runtime, workflow orchestrator, local cache

Workflows

CLI and native macOS operations

Define your model stack once and reuse it across CLI and UI operations.

Built for local operators.

Melix is designed for developers and AI builders who operate local models daily—from CLI workflows to native macOS surfaces and menubar-level control.

CLI

Scriptable command flows

macOS App

Native surface, session visibility, and status monitoring

Menubar

Quick controls and operation snapshots

Start from the repo.

Requirements

Apple Silicon Mac
Supported macOS version
Swift
Python
uv

Quick Start

Clone the repository.
Install dependencies.
Start the local service.
Open the app or run CLI workflow.
Run your first model / benchmark / evaluation.

Start in 3 minutes

Copy and run the commands below.

git clone https://github.com/Keith-CY/melix.git
cd melix
cat README.md

Run checks

Run your first local checks

Run these commands after cloning to verify your environment.

swift --version
python --version
uv --version

Current Status

v0.1.0-alpha · developer preview

v0.1.0-alpha / Available now

v0.1.0-alpha · local model workflow, LoRA, benchmark/eval, CLI and macOS surfaces

v0.2.0-beta / In progress

v0.2.0-beta · stronger workflow state, agent/tool orchestration, long-running automation

v0.3.0 / Not yet

v0.3.0 · production-grade autonomous agent orchestration, cloud collaboration, hosted service

Deployment source main · Updated · Deployment id n/a