PS5LM

Your PS5 runs the language model Every device on your network uses it

llama.cpp on the console's GPU, with OpenAI and Anthropic APIs and a screen made for the TV. No PC, no cloud, no account.

Measured on the console

A PS5 Slim, every layer on its GPU, with 11.4 GB of the shared memory for the model and its context

Swipe the table to see the speeds

ModelSizeOn a PS5 Slim
Qwen 3.5 0.8B, Q4_K_M0.5 GB207 tokens a second
Qwen 3.5 9B, Q4_K_M5.3 GB55 tokens a second
Qwen 3.8 27B, UD-Q2_K_XL9.2 GBabout 20 tokens a second, 64k context
d1 Omni 600M, a decision model1.1 GB3 questions in 0.3 s

A 720p image and seven questions take d1 Omni 600M 2.9 s. Generation is bound by memory bandwidth: the 9B model reads its weights at about 70% of the GPU's 448 GB/s.

Made for the TV

Five pages you move between with L1 and R1, all in reach of the DualSense

Home: the model, its speed, and the requests as a chat
HomeThe model, its speed, and every request and its answer as a chat
Models: suggested models measured on the console
ModelsWhat is on this PS5, models measured on the console, Hugging Face search
Home with a decision model loaded
DecisionsWith a decision model, Home shows the time per decision and each answer
Console: temperature, power, GPU and CPU charts
ConsoleTemperature, power, GPU and CPU over two minutes or an hour
API: the endpoints and an example
APIThe address, the endpoints and a request to copy
Settings with a list of choices under a row
SettingsStartup, context and cache, storage, and the logs

Use it from any device

Once Home says Ready, the console answers on port 8081. The model name in a request is ignored: the loaded one answers.

1Chat OpenAI API
# the address Home shows on the TVPS5=http://192.168.8.53:8081 curl $PS5/v1/chat/completions \  -H 'Content-Type: application/json' \  -d '{"model": "ps5", "messages": [    {"role": "user",     "content": "Hello from my laptop"}]}'
2Decide /v1/systemone
curl $PS5/v1/systemone \  -H 'Content-Type: application/json' \  -d '{"state": "Charged twice. Refund me.",  "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?"}}}'

Decision models are powered by Ollaya, which runs them on a PC the way Ollama runs language models: one forward pass, no text generated. The Suggested list has d1 Omni 600M, d1 3B and Clef Flash. Claude Code, opencode and codex setups are in docs/CLI.md.

Install

A PS5 on firmware 7.00 to 13.60, jailbroken, with these payloads loaded

PayloadForNeeded
kstuff-liteThe kernel patches homebrew runs onRequired
etaHENHomebrew supportRequired
ShadowMountPlusPuts PS5LM on the home screen from /data/homebrewRequired
elfldrTemperature, power and real free space; full-speed downloads, through the app's own ftpsrvRecommended
ftpsrvInstalling the app, copying models from a PC; downloads write through it when it runsRecommended
  1. 1Downloadps5lm-app-vX.Y.Z.zip from the latest release, and unzip it
  2. 2CopyThe PPSA99575 folder to /data/homebrew over FTP. ShadowMountPlus adds PS5LM to the home screen
  3. 3Get a modelIn Models > Suggested, or copy a .gguf to /data/PS5LM/models or to PS5LM/models on a USB drive
  4. 4Load itOnce Home says Ready, open http://<console>:8081 in a browser, or point your client at it

To update, copy the new folder over the old one: settings, models and downloads stay in /data/PS5LM. If the app says it cannot write to /data/PS5LM, the HEN has not opened storage to it; the fix for each HEN is in the README.

Questions

What people ask before they install it

Run it on your PS5

Free and open source, for a jailbroken PS5 you own on firmware 7.00 to 13.60

Get the app