PS5LM

// Overview

PS5LM runs llama.cpp on the GPU of a jailbroken PlayStation 5 and serves GGUF language models to the other devices on your network. While it runs there is no PC, no cloud service and no account involved: the console downloads models from Hugging Face, loads them, and answers requests on port 8081 through an OpenAI-style /v1/chat/completions endpoint, an Anthropic-style /v1/messages endpoint and a browser chat page.

The GPU work goes through llama.cpp's Vulkan backend on the console's RDNA 2 graphics. The model and its context share 11.4 GB of the PS5's memory, and a model planner picks the longest context that fits the free memory. The project's own measurements on a PS5 Slim are 207 tokens/s for Qwen 3.5 0.8B, 55 tokens/s for Qwen 3.5 9B and about 20 tokens/s for Qwen 3.8 27B at a 2-bit quantization with a 64k context; the project says generation is limited by memory bandwidth.

On the TV, PS5LM shows the incoming requests, a model library, an API page and a Console page with temperature, power and load. It also serves decision models from Ollaya through a /v1/systemone endpoint, and reads images with models that ship an image projector, such as Gemma 4 and MiniCPM-V. It needs a PS5 you own on firmware 7.00 to 13.60 with a jailbreak and the kstuff-lite, etaHEN and ShadowMountPlus payloads, and is released under GPL-3.0.

// What it does

• Runs llama.cpp on the PS5 GPU through its Vulkan backend, with any GGUF model that fits in 11.4 GB of shared memory
• OpenAI /v1/chat/completions and Anthropic /v1/messages endpoints plus a browser chat page on port 8081
• Ollaya-compatible decision models through /v1/systemone
• Image input for models with an image projector, such as Gemma 4, MiniCPM-V and d1
• Downloads models from Hugging Face on the console and stores them internally or on an M.2 or USB drive

• A TV interface with Home, Models, API, Console and Settings screens; the Console page tracks temperature, power and load

// Getting started

PS5LM needs a PS5 you own on firmware 7.00 to 13.60, jailbroken, with the kstuff-lite, etaHEN and ShadowMountPlus payloads loaded. elfldr and ftpsrv are recommended for telemetry and copying files.

curl http://<console>:8081/v1/systemone -H 'Content-Type: application/json' -d '{
  "state": "I was charged twice this month. Refund me or I cancel.",
  "questions": {
    "intent": {"type": "choice", "instructions": "What does the customer want?",
               "criteria": {"refund": null, "information": null, "cancel": null}},
    "angry":  {"type": "noul", "instructions": "Is the customer angry?"}
  }
}'

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// How PS5LM compares

PS5LM alongside other open-source local runtimes tools AI/TLDR tracks, ranked by GitHub stars.