About
UnKenny is a module for Foundry Virtual Tabletop. It allows you to generate a tiny dialogue with a character, by providing them with a limited AI. The idea arose during our Starfinder game, because we wanted to simulate the interaction with a robotical NPC.
Security
Be mindful which modules you enable.
If you are storing your OpenAI API key in UnKenny, you are entrusting it to us, and by extension, to FoundryVTT. The modularity of Foundry is one of its big selling points, but it is also an invitation to arbitrary code execution. Any module you install could, in principle, inject some code that steals your API key. Foundry takes some countermeasures like scoping of modules, but ultimately it is very hard to defend against attacks from within your application.
For that reason, monitor your API key usage every now and then. If ever you find a module that has stolen your key, please
Usage
Global Configuration
After enabling the module for a world, configure it in the global settings. Selecting a Large Language Model is required. If you want to use one of OpenAI's models, you will also need to set up and pay for an API key. Note that different models may have different capabilties and also different usage costs.
The other parameters are set to reasonable defaults.
Actor Configuration
Open the sheet of the actor that you want to make UnKenny and click on "Modify UnKennyness".
Provide the following two inputs:
- Alias: An UnKenny alias to refer to the actor in chat messages. Only alphanumeric characters and underscores are allowed here.
- Preamble: This is a short description of what the character needs to know about themselves.
You may also overwrite any global settings for this actor only. If for example you want to have a particularly repetitive character, you could set the Repetition penalty to a negative value. However, most effects can also be achieved by accurately describing the character in the preamble.
As long as the alias is not empty, the actor is considered unkenny and can be adressed. To speak to the actor, mention them anywhere in your message using @<alias>, where <alias> is their UnKenny alias.
Your chat message is posted, and a short while later, the response generated by the large language model is posted below it. This example was generated with the GPT-4 model.
Any messages previously posted at or by the actor are included in the chat conversation. Note that the context size of a large language model is limited. At some point UnKenny might warn you that it starts truncating messages. To avoid that, simply clean up the chat log. Alternatively you can switch to a model with a larger context size. For OpenAi, these are usually the ones ending on 'turbo'.