PS5LM is an open project that turns a jailbroken PS5 into a private model server at home. It runs llama.cpp on the console's GPU, loading any GGUF model and serving it to laptops, phones and coding tools on your network. No PC, cloud or account is required while it runs.
The console exposes OpenAI's /v1/chat/completions and Anthropic's /v1/messages APIs, so Claude Code can work against it, and a chat page works in any browser. The TV interface shows each incoming request as a chat, including which device asked, what it asked, and the streamed reply. A Console page charts temperature, power, GPU and CPU load over the last two minutes or hour.
Decision models are also supported. Loading a model such as d1 Omni 600M lets the console answer Ollaya's /v1/systemone API, returning typed choices, scores and yes-or-no answers with calibrated probabilities in a fraction of a second. Models with image projectors, including Gemma 4, Mini CPM-V and d1, can process pictures in chat and decisions.
Performance varies by model. On a PS5 Slim, Qwen 3.5 9B (Q4_K_M, 5.3 GB) reaches about 55 tokens a second, while the smaller Qwen 3.5 0.8B runs at 207 tokens a second. The built-in planner reads each GGUF and selects the longest context and best cache that fit available memory. Setup requires a jailbroken PS5 on firmware 7.00 to 13.60 with the listed payloads, after which the app can download models from Hugging Face directly to internal storage, the M.2 drive or a USB drive.
Source: Hacker News · Summarized by HeadlinesBriefing