ds4.pinokio Slots a Full DeepSeek V4 Brain Into Apple Silicon Macs With One Click

    
        By vramkickedin    
     | 
    
            May 23, 2026 at 11:04 pm        
    
     | 
    
        2 min read

ds4.pinokio is a new launcher and browser interface that brings the massive DeepSeek V4 Flash AI model to Apple Silicon Macs. It builds on the ds4.c Metal-only inference engine created by antirez, making setup as simple as a single click when using Pinokio. The project also includes a standalone web chat UI and local proxy that lets you interact with the model through an OpenAI-compatible API.

Developer cocktailpeanut wrapped the command-line ds4.c server and models into a user-friendly package. He added a Node-based web frontend that stores conversations locally and handles streaming reasoning output. This solves the problem of needing a quick, graphical way to test and use the large model without complex terminal commands.

High-memory hardware required

Key Features

One-click install via Pinokio launcher.
Built-in web UI with local chat history.
Streaming output for reasoning steps.
Download q2 (81GB) or q4 (153GB) models.
Manual install without Pinokio dependency.
Optional MTP draft model speeds up generation.
Disk KV cache reduces memory pressure.
Local proxy keeps API calls on-device.

This tool targets owners of high-end Apple Silicon Macs with at least 128GB of unified memory, such as the M3 Ultra. Privacy-conscious professionals can run the massive model entirely offline for sensitive tasks. Hobbyists and small agencies looking to experiment with local AI reasoning models will find the one-click setup especially convenient.

Project notes and limitations

The web UI is a purpose-built, lightweight chat surface that does not rely on llama.cpp's server interface, instead using its own minimal Node.js proxy. A major constraint is that ds4.c only works with the specific DeepSeek V4 Flash GGUF files and requires Metal, so it is limited to Apple Silicon Macs. The developer notes that the q2 model alone needs about 81GB of disk space and 128GB of RAM, making it inaccessible for most consumer hardware.