AI Backends - guspuffygit/sentient-sims-app GitHub Wiki

There are currently three options to use as an AI backend to play Sentient Sims.

OpenAI

Open AI is the service behind ChatGPT. The mod uses the Open AI API and uses the same model that runs ChatGPT.

You will need to create or have an account with OpenAI in order to use the mod.

Creating an API Key with Open AI allows the mod to use the API to generate just like ChatGPT but inside the Sims.

Your API key is a secret and is never sent to Sentient Sims and stays local to your computer.

Setup OpenAI

OpenAI Free Trial

OpenAI offers a free trial of their API, but the limits on it are so restrictive that it makes the mod unplayable. You can try this, but you will get errors.

OpenAI Pay-as-you-go cost

Players of the mod have reported very low cost per month when playing using OpenAI. It is usually $1-$3 a month even for heavy players.

In order to prevent yourself from spending too much money, set a SOFT/HARD spending limit on your account. This will prevent you from going over a certain dollar amount on accident. Most players set their HARD limit to $5 and never hit it.

A SOFT limit will notify you via email once you hit that amount.

A HARD limit will throw errors and stop you from making more requests once you hit this amount to prevent you from overspending on accident.

Set your soft/hard limit here: https://platform.openai.com/account/billing

Sentient Sims Uncensored AI

The Sentient Sims Uncensored AI was a service created to provide an uncensored and privacy focused AI service to use with the Sentient Sims mod.

Players sometimes find that ChatGPT/OpenAI has overbearing censorship and will refuse to write about certain scenarios. In addition, OpenAI will use or look at data sent to them to train or ban players if their requests go against the OpenAI terms of service.

To use the Sentient Sims Uncensored AI you will need to subscribe via Patreon. To setup look at

Local/Remote AI (Self Hosted/Free)

If you would like to run the Sentient Sims Uncensored AI locally, you can do so if you have the proper hardware to do so. Easiest and most recommended is using an RTX 3090/4090 GPU. This large GPU is required because of how AI inference works using large amounts of VRAM.

Sentient Sims is compatible with KoboldAI API backends, so any version of KoboldAI (United-Branch/0cc4m/Koboldcpp) can be used as a backend.

Koboldcpp can be used to offload some of the model to your CPU/memory instead of all in the GPU in case you have a smaller GPU. This makes the model incredibly slow to respond so your mileage may vary.

The current model used with Sentient Sims is MythoMax-L2-13B-GPTQ. Other models could be used in place of this, but this requires code changes to the sentient-sims-app to change the prompt formatting. I am constantly testing different models and this model has been the best so far to use.

Other AIs

There are many other services that could be used as an AI backend. I have investigated novelai and some others to be added as options but I haven't added them yet.

If you want to contribute and add an AI backend to the mod, the app that handles the requests is open source and I welcome contributions!

If you have questions about contributing please visit the Discord.