cobanov/PS5LM: llama.cpp on a jailbroken PS5: run any GGUF mannequin natively on the console, beginning with Qwen 3.8 · GitHub

Your PS5 runs the language mannequin. Every system in your community makes use of it.
llama.cpp on the console’s GPU, with OpenAI and Anthropic APIs and a display made for the TV.
Get the app ·
Website ·
Get started ·
Issues
New: determination fashions, powered by Ollaya. PS5LM now runs Jev-like
determination fashions (d1, Clef, OpenJev and extra) on the GPU and solutions the identical/v1/systemone
API as Ollaya: typed selections with calibrated possibilities, in a fraction of a second.
PS5LM is for anybody with a jailbroken PS5 who needs a personal mannequin server at residence. It masses a GGUF
mannequin onto the console’s GPU and serves it to your laptop computer, your cellphone and your coding instruments. It wants
no PC whereas it runs, no cloud and no account.

Qwen 3.5 9B answering requests from a laptop computer on a PS5 Slim. Click for the video.






Home, Models, Console, API and Settings on the TV.
- It runs on the GPU. llama.cpp’s Vulkan backend on the console’s RDNA 2 GPU, with 11.4 GB of the
shared reminiscence for the mannequin and its context. Every structure and quantization llama.cpp is aware of
comes alongside. - It solutions your community. OpenAI’s
/v1/chat/completions, Anthropic’s/v1/messages(Claude Code
works in opposition to it) and a chat web page in any browser. Point a shopper athttp://and use:8081
any mannequin identify: the loaded one solutions. - It decides, too. Load a choice mannequin and the console solutions TypeProtected’s
/v1/systemone:
selection, rating and yes-or-no questions on textual content and pictures, the API Ollaya serves on a PC. - It reveals its work. Every request streams throughout the TV as a chat: which system requested, what it
requested, and the reply as it’s written. Square sends a demo and Triangle opens a keyboard, so the
console will be tried from the sofa. - It watches the console. A Console web page charts temperature, energy, GPU and CPU load over the
final two minutes or the final hour, beside a desk of each studying. - It picks what matches. The planner reads every GGUF and offers it the longest context and greatest cache
that match the reminiscence free. A Suggested record holds fashions measured on the console, and the app
downloads from Hugging Face by itself, to inside storage, the M.2 drive or a USB drive. - It reads pictures. Models with a picture projector (Gemma 4, MiniCPM-V, d1) take footage in chat
and in selections.
| Model | Size | On a PS5 Slim |
|---|---|---|
| Qwen 3.5 0.8B, Q4_K_M | 0.5 GB | 207 tokens a second |
| Qwen 3.5 9B, Q4_K_M | 5.3 GB | 55 tokens a second |
| Qwen 3.8 27B, UD-Q2_K_XL | 9.2 GB | about 20 tokens a second, 64k context |
| d1 Omni 600M, a choice mannequin | 1.1 GB | 3 questions in 0.3 s; a 720p picture and seven questions in 2.9 s |
Ollaya runs open determination fashions on a PC the best way Ollama runs
language fashions. PS5LM brings the identical thought to the console: a choice mannequin reads a state and
solutions typed questions in a single ahead move, with no textual content generated.
curl http://<console>:8081/v1/systemone -H 'Content-Type: utility/json' -d '{
"state": "I used to be charged twice this month. Refund me or I cancel.",
"questions": {
"intent": {"kind": "selection", "directions": "What does the client need?",
"standards": {"refund": null, "info": null, "cancel": null}},
"indignant": {"kind": "noul", "directions": "Is the client indignant?"}
}
}'
The Suggested record has d1 Omni 600M, d1 3B and Clef Flash. Any GGUF that carries llama.cpp’s
determination metadata masses the identical manner, and the app labels it a choice mannequin.
You want a PS5 on firmware 7.00 to 13.60, jailbroken, with these payloads loaded:
| Payload | For | |
|---|---|---|
| kstuff-lite | the kernel patches homebrew runs on | required |
| etaHEN | homebrew assist | required |
| ShadowMountPlus | places PS5LM on the house display from /knowledge/homebrew |
required |
| elfldr (port 9021) | temperature, energy and actual free house, by means of ps5-exporter | really useful |
| ftpsrv (port 2121) | putting in the app, copying fashions from a PC | really useful |
- Download
ps5lm-app-vX.Y.Z.zipfrom the latest release and unzip it. - Copy the
PPSA99575folder to/knowledge/homebrewover FTP. ShadowMountPlus provides PS5LM to the
residence display. - Open PS5LM, go to Models > Suggested and get a mannequin. Or copy a
.ggufto
/knowledge/PS5LM/fashions, or toPS5LM/fashionson a USB drive. - Load it. Once Home says Ready, open
http://in a browser, or level your OpenAI or:8081
Anthropic shopper at it.
To replace, copy the brand new folder over the outdated one. Settings, fashions and downloads keep in
/knowledge/PS5LM. Using it from coding instruments (opencode, Codex, Claude Code) is in
docs/CLI.md.
| Where | Button | Does |
|---|---|---|
| Everywhere | L1 / R1 | Home, Models, API, Console, Settings |
| Home | Cross | change the mannequin |
| Home | Square / Triangle | ship a demo request / kind a message |
| Models | D-pad, Cross | select a bit and a mannequin; load, unload or obtain it |
| Models, Search | L2 / R2, Square | change the search; verify whether or not a file matches |
| Models, Downloads | Cross, Square | pause or resume; cancel (press twice) or clear |
| Console | L2 / R2 | the final two minutes / the final hour |
| Settings | D-pad, Cross | select a class and a row; flip it, or decide from its record |
| Settings, Diagnostics | Square, Triangle | app or server log; errors solely |
C++ on ps5-homebrew-ui, llama.cpp with
ggml-vulkan on PS5_Vulkan‘s RADV, constructed on Linux.
Building · Console notes · Porting ·
Research · Models that fit · Roadmap ·
Changelog · The original payload
GPL-3.0-or-later, see LICENSE. llama.cpp retains its MIT licence, Manrope its OFL. No Sony
SDK information, keys, firmware or mannequin weights are on this repository.
PS5LM will not be affiliated with Sony Interactive Entertainment. “PlayStation” and “PS5” are emblems
of Sony Interactive Entertainment.
