Keith-CY/melix

Local-first AI runtime for Apple Silicon with CLI and macOS operator workflows for LoRA training, benchmarking, and evaluation.

7

stars

2,930

commits

Swift

primary language

Sep 7, 2026

updated

www.melix.run
ai-runtime
apple-silicon
inference-server
local-ai
local-llm
lora
lora-training
macos
mlx
mlx-lora
model-ops
qlora
swift
Browse cluster: MLX framework for Apple Silicon ML

README

Melix logo

Your Mac. Your Models. Your Rules.

A native local AI workspace for Apple Silicon — run, fine-tune, and benchmark language models
entirely on your own machine, with no data leaving your desk.

release gates app packaging platform macOS focus license


What Is Melix?

Melix is a complete local AI operations platform built for Apple Silicon Macs. It brings together everything you need to work with language models — serving, fine-tuning, benchmarking, and evaluation — in a single native app and CLI, with no subscriptions, no cloud dependencies, and no data leaving your machine.

Think of it as a local AI studio: a place where you can load a model, adapt it to your needs, measure its quality, and compare results — all from your own hardware.

Melix Window UI showing a LoRA-derived local server


What You Can Do With Melix

CapabilityWhat It Means For You
Model RegistryImport models from disk or download from Hugging Face into a local registry you control
Server SessionsStart, pause, resume, and stop local model servers with a single command or click
Local ChatChat with any registered model from the menubar app or CLI, staying fully offline
LoRA Fine-TuningTrain a custom adapter on your own dataset to specialize a model for your use case
BenchmarkingRun repeatable quality benchmarks and matrix comparisons across models and adapters
EvaluationScore models against standard suites (MMLU and more) and export the results
Native macOS AppA polished menubar and workspace UI that puts all of the above one click away

Why Melix?

Local AI work tends to scatter across too many tools: one script for serving, another for training, a notebook for evaluation, and a shell history for benchmarks. Melix keeps that entire loop in one place, on one machine.

  • Privacy by default. Your models, your data, your results — none of it leaves your Mac.
  • No subscriptions. Run any compatible model as many times as you like at zero marginal cost.
  • Repeatable science. Benchmark and evaluation results are stored in a repository-owned format so comparisons stay honest over time.
  • Operator-grade UI. The native macOS workspace is a first-class citizen, not an afterthought.

Who Is Melix For?

  • AI practitioners on Apple Silicon who want a local runtime instead of a remote-only workflow
  • Model engineers who need a repeatable loop for LoRA fine-tuning, comparison, and evaluation
  • Privacy-conscious builders who want full control over inference without cloud APIs
  • Local AI enthusiasts who want a polished native macOS experience alongside the CLI
  • Contributors who care about typed protocols, reproducible runbooks, and productized local tooling

Quick Start

Make sure you have macOS 15+ on Apple Silicon, plus swift, python3, and uv installed.

# 1. Bootstrap the repository
make bootstrap

# 2. Generate protocol artifacts
make proto

# 3. Run verification gates
make swift-test
make py-test
make integration-test

Then bring up the local stack:

bash scripts/dev_up.sh

For the full native macOS experience:

bash scripts/dev_app_up.sh

New here? The Getting Started guide walks through every step with context.


Documentation

DocumentWhat's Inside
Getting StartedStep-by-step setup from a fresh checkout to a working loop
Current StatusWhat is shipped today and where the honest boundaries are
Phase RoadmapThe original phase model and its current completion state
ContributingHow to contribute, what to verify, and what a good PR looks like
Docs MapFull index of specs, runbooks, architecture decisions, and plans
Marketing & Storytelling KitProduct overview, LoRA narrative, reusable copy, and screenshot evidence
Benchmark & LoRA RunbookDeep-dive operator guide for benchmarking and fine-tuning
Local Install RunbookInstall Melix as a persistent local service
Packaging TargetsHomebrew, launch agent, and app-bundle delivery options

Contributing

Contributions are welcome — documentation, tooling, protocol, runtime, CLI, and macOS operator surface. Start with docs/contributing.md for the workflow, verification commands, and handoff expectations.

The short version:

  1. Branch from main.
  2. Keep the change small and focused on one behavior slice.
  3. Update the relevant spec or runbook if behavior changes.
  4. Run make swift-test && make py-test && make integration-test before opening a PR.

License

Melix is licensed under the Apache License, Version 2.0. See LICENSE.

Contributors

Zigfreidish

2,359 commits

Keith-CY

316 commits

Akane-CN

237 commits

dependabot[bot]

18 commits

Keith-CY/melix

Local-first AI runtime for Apple Silicon with CLI and macOS operator workflows for LoRA training, benchmarking, and evaluation.

7

stars

2,930

commits

Swift

primary language

Sep 7, 2026

updated

www.melix.run
ai-runtime
apple-silicon
inference-server
local-ai
local-llm
lora
lora-training
macos
mlx
mlx-lora
model-ops
qlora
swift
Browse cluster: MLX framework for Apple Silicon ML

README

Melix logo

Your Mac. Your Models. Your Rules.

A native local AI workspace for Apple Silicon — run, fine-tune, and benchmark language models
entirely on your own machine, with no data leaving your desk.

release gates app packaging platform macOS focus license


What Is Melix?

Melix is a complete local AI operations platform built for Apple Silicon Macs. It brings together everything you need to work with language models — serving, fine-tuning, benchmarking, and evaluation — in a single native app and CLI, with no subscriptions, no cloud dependencies, and no data leaving your machine.

Think of it as a local AI studio: a place where you can load a model, adapt it to your needs, measure its quality, and compare results — all from your own hardware.

Melix Window UI showing a LoRA-derived local server


What You Can Do With Melix

CapabilityWhat It Means For You
Model RegistryImport models from disk or download from Hugging Face into a local registry you control
Server SessionsStart, pause, resume, and stop local model servers with a single command or click
Local ChatChat with any registered model from the menubar app or CLI, staying fully offline
LoRA Fine-TuningTrain a custom adapter on your own dataset to specialize a model for your use case
BenchmarkingRun repeatable quality benchmarks and matrix comparisons across models and adapters
EvaluationScore models against standard suites (MMLU and more) and export the results
Native macOS AppA polished menubar and workspace UI that puts all of the above one click away

Why Melix?

Local AI work tends to scatter across too many tools: one script for serving, another for training, a notebook for evaluation, and a shell history for benchmarks. Melix keeps that entire loop in one place, on one machine.

  • Privacy by default. Your models, your data, your results — none of it leaves your Mac.
  • No subscriptions. Run any compatible model as many times as you like at zero marginal cost.
  • Repeatable science. Benchmark and evaluation results are stored in a repository-owned format so comparisons stay honest over time.
  • Operator-grade UI. The native macOS workspace is a first-class citizen, not an afterthought.

Who Is Melix For?

  • AI practitioners on Apple Silicon who want a local runtime instead of a remote-only workflow
  • Model engineers who need a repeatable loop for LoRA fine-tuning, comparison, and evaluation
  • Privacy-conscious builders who want full control over inference without cloud APIs
  • Local AI enthusiasts who want a polished native macOS experience alongside the CLI
  • Contributors who care about typed protocols, reproducible runbooks, and productized local tooling

Quick Start

Make sure you have macOS 15+ on Apple Silicon, plus swift, python3, and uv installed.

# 1. Bootstrap the repository
make bootstrap

# 2. Generate protocol artifacts
make proto

# 3. Run verification gates
make swift-test
make py-test
make integration-test

Then bring up the local stack:

bash scripts/dev_up.sh

For the full native macOS experience:

bash scripts/dev_app_up.sh

New here? The Getting Started guide walks through every step with context.


Documentation

DocumentWhat's Inside
Getting StartedStep-by-step setup from a fresh checkout to a working loop
Current StatusWhat is shipped today and where the honest boundaries are
Phase RoadmapThe original phase model and its current completion state
ContributingHow to contribute, what to verify, and what a good PR looks like
Docs MapFull index of specs, runbooks, architecture decisions, and plans
Marketing & Storytelling KitProduct overview, LoRA narrative, reusable copy, and screenshot evidence
Benchmark & LoRA RunbookDeep-dive operator guide for benchmarking and fine-tuning
Local Install RunbookInstall Melix as a persistent local service
Packaging TargetsHomebrew, launch agent, and app-bundle delivery options

Contributing

Contributions are welcome — documentation, tooling, protocol, runtime, CLI, and macOS operator surface. Start with docs/contributing.md for the workflow, verification commands, and handoff expectations.

The short version:

  1. Branch from main.
  2. Keep the change small and focused on one behavior slice.
  3. Update the relevant spec or runbook if behavior changes.
  4. Run make swift-test && make py-test && make integration-test before opening a PR.

License

Melix is licensed under the Apache License, Version 2.0. See LICENSE.

Contributors

Zigfreidish

2,359 commits

Keith-CY

316 commits

Akane-CN

237 commits

dependabot[bot]

18 commits

Languages

Swift

54.0%

Python

45.8%