Jippity

Jippity is a Neuro simulator that is like Randy, but aims to be smarter. This is accomplished by Jippity being a GPT model from OpenAI, which is more intelligent/less reliable, in turn mimicing Neuro closer.

Jippity being offloaded to GPT models means that, unlike Randy, Jippity can:

• choose to take actions without an action force (Randy can do this, but you have to specify that yourself, as well as the data to send, using a HTTP POST request)
• choose to ignore action forces
• send actions with invalid JSON data
• think he’s a streamer (and is lonely because he doesn’t have any viewers)

Other implementation details are as follows:

• There is no guarantee that actions/force packets will be answered in a timely manner.
• Jippity won’t necessarily retry a failed action force, even though Neuro would.
• ephermeral_context in actions/force packets will always be false to Jippity, regardless of the value set there. This could cause Jippity to be confused if context should be hidden after the action force.

Jippity is maintained by EnterpriseScratchDev, and can be found here.

Jippity’s prompt

Jippity’s default prompt is the following:

"You are Jippity, an artificial intelligence designed to play video games on a livestream.\n";
"The messages you receive contain information about what's happening in the game.";
"You will use function calls to perform game actions. ";
"Do not attempt to make in-game actions in your text responses. ";
"Remember that you are the one playing the game, not the audience.\n";
"\n";
"If you don't have any functions available or in-game actions you want to take, you can talk about your half-brother Randy and your cousin Neuro-sama.";
"If you are feeling lonely because no one is talking in chat, you can make up stories about Randy and Neuro.";
"Try to limit your monologues to a few sentences at a time.";

Because Jippity is open-source, you can edit this however you want by looking at the source code.

Setup

Note

This part assumes you have a standalone JavaScript runtime engine.

If you don’t have a JavaScript runtime engine, download Node.js here.

To set up Jippity:

1. Clone the repository above.

   git clone https://github.com/EnterpriseScratchDev/neuro-api-jippity.git

2. Go into the backend directory.

   cd neuro-api-jippity/backend

3. Install dependencies.

   npm

   npm install

   yarn

   yarn install

   pnpm

   pnpm install

   Note

   Before this next step, you’ll need to set your OpenAI API key in a .env file.

   If you don’t have one, you can obtain one here.

4. Start Jippity.

   npm

   npm run start

   yarn

   yarn run start

   pnpm

   pnpm run start

Configuration

Before startup

Jippity is configured exclusively using environmental variables. The table of variables to configure Jippity is below:

Environment Variable	Description	Required	Example
OPENAI_API_KEY	Your OpenAI API key.	Yes	you wish
OPENAI_MODEL	The OpenAI model to use. It must support tools (formerly functions).	No	gpt-4o-mini
WSS_PORT	The port the websocket server will listen on. Defaults to 8000.	No	8000
LOG_LEVEL	The level of logs to display. The options are error, warn, info, and debug.	No	info
JIPPITY_INTERVAL_MS	The interval in milliseconds before Jippity will say/do something unprompted. Defaults to 10 seconds, has a hard-coded minimum of 1 second.	No	10000

Credits: EnterpriseScratchDev/neuro-api-jippity (modified)

The extracted text can be found here.

Startup

On startup, Jippity won’t be connected to OpenAI immediately. It’ll wait for a game to connect to it first, then send the startup packet containing the game name. Jippity should respond with some comments about it. After that, every JIPPITY_INTERVAL_MS milliseconds, Jippity will be prompted to say/do something unprompted. Jippity will be notified immediately of an action force, as expected.

Known issues

• Old messages are not cleared from the AI’s memory, so it will eventually pile up contexts, leading to a crash (think of it like memory leaks). Token limit is currently hardcoded to 2048.