Claude Code Local runs models on-device on Apple Silicon via a local MLX native server, interfacing with Claude Code without cloud calls. It supports Gemma 4 31B, Llama 3.3 70B, Qwen 3.5 122B, and DeepSeek via ds4, with 65 tok/s speed for Qwen 3.5 and local privacy guarantees.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Claude Code Local is a local, on-device MLX-based server that runs Claude Code models on Apple Silicon without internet access or API keys. It exposes a local API that Claude Code uses directly, bypassing cloud proxies.
How it works
The repository provides a single proxy-like server file (~1000 lines) that loads models via Apple's MLX/Metal stack, translates tool calls between local models and Claude Code, cleans and caches prompts, and optimizes prompt prefill to reduce latency. It supports multiple models (Gemma 4 31B, Llama 3.3 70B, Qwen 3.5 122B, and DeepSeek V4 via ds4) and can run with a local server wrapper or launcher scripts.
Getting started
git clone https://github.com/nicedreamzapp/claude-code-local
cd claude-code-local
bash setup.sh
The setup script auto-detects RAM, downloads a model, installs the MLX server, and creates a Claude Local.command launcher on the Desktop.
Manual start example:
# 1. Set up the MLX virtualenv
python3.12 -m venv ~/.local/mlx-server
~/.local/mlx-server/bin/pip install mlx-lm
# 2. Pick a fighter and download
bash scripts/download-and-import.sh gemma
# 3. Start the server
MLX_MODEL=divinetribe/gemma-4-31b-it-abliterated-4bit-mlx \
bash scripts/start-mlx-server.sh
# 4. Launch Claude Code
ANTHROPIC_BASE_URL=http://localhost:4000 \
ANTHROPIC_API_KEY=sk-local \
claude --model claude-sonnet-4-6
Recent releases
- v0.1.0 — verified-offline + 16GB Mac support (2026-05-08): First tagged release. This locks in two big shifts that landed this week.
Traction
3078 stars, 588 forks, 3 open issues
Behind the repo
The project is presented as an MLX-native server with local models and an Anthropic API interface. The README describes the server as a single file (~1000 lines) and mentions the server path at ~/.local/mlx-native-server/server.py.
Caveats
License: MIT. Created 2026-03-26; last push 2026-07-19. Requires Python 3.12+ and Claude Code CLI (npm install -g @anthropic-ai/claude-code). Model downloads range ~8-75 GB depending on fighter. Abliteration notes and licensing for upstream models apply.