🗣️ Yapping Agents

• Yapping Agents is a simple program that demonstrates multiple AI agents continuously chat with each other.
• This program runs LLM locally so you don't need to rely on external api calls.

😇 Thanks to llama.cpp for providing the tools to run the LLM agents locally.

🤔 What Does It Do?

• Agents with Persona (LLM personas) continuously chat with each other.
• The RepeatingChecker agent detects any repetitive loops which is a common issue with AI agents.
• The RandomTopicGenerator agent provides randomly generated but unique topics to keep the conversation fresh.

🏗️ How to Use

1. Clone this repository:

   git clone git@github.com:Honeybeei/yapping-agents.git

2. Install dependencies:

   cd yapping-agents
   npm install

3. Download a GGUF model of your choice and place it into the ./gguf-models directory.

   • Example using unsloth/DeepSeek-R1-Distill-Llama-8B-GGUF:

     npx node-llama-cpp pull --dir ./gguf-models https://huggingface.co/unsloth/DeepSeek-R1-Distill-Llama-8B-GGUF

⚠️ Note: Change the modelName in the config.json file to match the model you downloaded.

4. Run the program:

   npm start

⚙️ Configuration

You can modify the settings in the config.json file:

{
  "modelName": "unsloth_DeepSeek-R1-Distill-Llama-8B-GGUF_DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf",
  "persona": {
    "temperature": {
      "min": 0.8,
      "max": 1.5
    },
    "personality": {
      "traitCount": 10
    },
    "prompts": {
      "system": ""
    }
  }
}

• modelName: Your chosen model file’s name in the ./gguf-models folder.
• temperature: Adjust the randomness and creativity of the chat.
  • Higher values produce more varied responses.
  • Lower values produce more predictable responses.
• traitCount: Number of personality traits assigned to each agent.
• system: Extra system prompt you can provide to shape the persona’s responses.