>>25354There's Oogabooga, which is the bare minimum open-source thing you can use, TavernAI is an offshoot but it's more for roleplay use and is extremely behind, Faraday is the most up-to-date and good for use with LLMs, but is primarily intended for roleplay
Locally hosted models also require a bit of compute power to run locally by the way, ChatGPT 3.5 is 107b parameters, with each billion requiring 1gb of VRAM, the rest can load on regular ram, but it's going to get slower as the models get larger