Run Claude Code Locally on Apple Silicon Using LM Studio and LiteLLM | Tech Edge AI

Running Claude Code Locally on Apple Silicon
📌 The method allows running Claude Code locally on Apple Silicon Macs (M1, M2, M3) using LM Studio and Light LLM at zero cost, eliminating recurring API fees.
💻 This local setup leverages Apple Silicon hardware for enhanced privacy and performance, completely offline, contrasting with cloud-based API usage.
⚙️ The architecture uses Light LLM as a translation layer to bridge the gap, as Claude Code expects the Anthropic Messages API while local LLMs use an OpenAI compatible API.

Prerequisites and Initial Setup
🛠️ Required prerequisites include an Apple Silicon Mac, macOS, Python 3.1+, Node.js, sufficient RAM for a 30B parameter model, and LM Studio installed.
📥 In LM Studio, download the Quinn 3 coder 30B model, load it, and enable the local server feature, noting the OpenAI compatible chat completions API endpoint.
🐍 Set up a clean Python virtual environment, activate it, and install Light LLM with proxy support using `pip`.

Configuration and Deployment
📝 Create a `config.yaml` file for Light LLM to map Claude Code model names to the actual LM Studio model ID and drop Anthropic-specific parameters.
📡 Launch the Light LLM proxy server, which acts as a bridge, translating requests from Claude Code for the local model running in LM Studio.
🧪 Verify communication between Light LLM and LM Studio using a cURL command to test the connection with the Quinn 3 coder 30B model.

Final Integration and Benefits
🌐 Install Claude Code globally (via npm or direct command) and set environment variables to point it toward the local Light LLM proxy, ensuring all requests bypass the cloud.
🚀 Start Claude Code, which will now operate using the local model, enabling it to perform complex tasks like reading multifile code bases, running tests, refactoring, and debugging securely and privately.
🐌 While local models might be slower than cloud versions, performance is generally good on Apple Silicon, and models like Quinn 3 coder 30B excel at tasks like test generation and large-scale repository changes.

Key Points & Insights
➡️ To achieve the best local performance on Apple Silicon, use LM Studio with MLX models (like Quinn 3 coder 30B), as OAMA does not fully utilize the hardware due to lack of MLX support.
➡️ The Light LLM proxy is the critical component that enables seamless communication by translating API protocols between Claude Code and local LLM runtimes.
➡️ Running the entire workflow locally ensures zero cost and complete data privacy for advanced Agentic Coding development.

📸 Video summarized with SummaryTube.com on Jan 25, 2026, 19:04 UTC