Aether v0.87 by helloitu
Runs local language models on the PS4 with llama.cpp and serves them over OpenAI- and Anthropic-style HTTP endpoints
Install on your PS4
Package (.pkg): with GoldHEN running, copy Aether-v0.87.pkg to the root of a USB drive (exFAT or FAT32), plug it into the PS4 and install it from Settings › Debug Settings › Game › Package Installer.
From a computer or phone: give this link to a remote package sender (GoldHEN's Remote Package Install, or Remote Package Installer on port 12800):
https://store.brewology.com/dl/6164/Aether-v0.87.pkg
Files
| File | Size | SHA-256 | |
|---|---|---|---|
| Aether-v0.87.pkg | 10.2 MB | c5d73351…c23b | Download |
Screenshots
About Aether
The app on the TV also shows the server log and lets you change token limits, temperature and the response mode. An experimental bridge hands some of the maths to the PS4's GPU.
Stick to small models for now. The author suggests Qwen 2.5 0.5B Instruct and says bigger ones, from around 2B, can crash during inference. Long prompts slow it down a lot on a base PS4.
Release notes
v0.87 · Jul 9, 2026
First release
Need to do bether:
-> DEBUG and fix crash when inference happens with bigger models (> 2B+)
-> Adapt APP type to get more VRAM alocation
-> Enable in-app video recording with native SHARE feature
Suggested GGUF to run inference: QWEN 2.5 0.5B Instruct
