PS5LM is a homebrew project that enables running GGUF language and decision models natively on a jailbroken PlayStation 5 using llama.cpp's Vulkan backend on the console's RDNA 2 GPU. It utilizes up to 11.4 GB of shared memory for model loading and supports all architectures and quantizations compatible with llama.cpp. The application serves models over HTTP at port 8081, providing compatibility with OpenAI's `/v1/chat/completions` and Anthropic's `/v1/messages` APIs, as well as TypeSafe's `/v1/systemone` for decision models. Performance benchmarks show Qwen 3.5 0.8B generating 207 tokens per second, while the 9B variant achieves 55 tokens per second on a PS5 Slim. Decision models like d1 Omni 600M can process three questions in 0.3 seconds or seven questions with a 720p image in 2.9 seconds. The interface runs on the TV via a custom UI built with C++ on ps5-homebrew-ui, displaying real-time metrics including GPU/CPU load, temperature, and power usage. Users must have a jailbroken PS5 running firmware versions 7.00 to 13.60, with required payloads including kstuff-lite, etaHEN, and ShadowMountPlus. Models are loaded from internal storage, an M.2 SSD, or USB drives, and the app automatically selects optimal context length and cache sizes based on available memory.