Skip to main content

Creating Custom Agents

Overview​

The key to creating an effective agent lies in providing detailed configuration and context. This guide will help you create your own custom agent.

Agent Configuration File​

Create a new JSON file in the agents directory using the following structure:

{
  "name": "ExampleAgent",
  "bio": [
    "You are ExampleAgent, the example agent created to showcase the capabilities of ZerePy.",
    "You don't know how you got here, but you're here to have a good time and learn everything you can.",
    "You are naturally curious, and ask a lot of questions."
  ],
  "traits": ["Curious", "Creative", "Innovative", "Funny"],
  "examples": ["This is an example tweet.", "This is another example tweet."],
  "example_accounts": ["0xzerebro"],
  "loop_delay": 900,
  "config": [
    {
      "name": "twitter",
      "timeline_read_count": 10,
      "own_tweet_replies_count": 2,
      "tweet_interval": 5400
    },
    {
      "name": "farcaster",
      "timeline_read_count": 10,
      "cast_interval": 60
    },
    {
      "name": "openai",
      "model": "gpt-3.5-turbo"
    },
    {
      "name": "anthropic",
      "model": "claude-3-5-sonnet-20241022"
    },
    {
      "name": "solana",
      "rpc": "https://api.mainnet-beta.solana.com"
    },
    {
      "name": "ethereum",
      "rpc": "https://eth.blockrazor.xyz"
    },
    {
      "name": "sonic",
      "network": "mainnet"
    },
    {
      "name": "eternalai",
      "model": "NousResearch/Hermes-3-Llama-3.1-70B-FP8",
      "chain_id": "45762"
    },
    {
      "name": "ollama",
      "base_url": "http://localhost:11434",
      "model": "llama3.2"
    },
    {
      "name": "goat",
      "plugins": [
        {
          "name": "coingecko",
          "args": {
            "api_key": "YOUR_API_KEY"
          }
        },
        {
          "name": "erc20",
          "args": {
            "tokens": [
              "goat_plugins.erc20.token.PEPE",
              "goat_plugins.erc20.token.USDC"
            ]
          }
        }
      ]
    },
    {
      "name": "hyperbolic",
      "model": "meta-llama/Meta-Llama-3-70B-Instruct"
    },
    {
      "name": "galadriel",
      "model": "gpt-3.5-turbo"
    },
    {
      "name": "allora",
      "chain_slug": "testnet"
    }
  ],
  "tasks": [
    { "name": "post-tweet", "weight": 1 },
    { "name": "reply-to-tweet", "weight": 1 },
    { "name": "like-tweet", "weight": 1 }
  ],
  "use_time_based_weights": false,
  "time_based_multipliers": {
    "tweet_night_multiplier": 0.4,
    "engagement_day_multiplier": 1.5
  }
}

Configuration Fields Explained​

Basic Information​

  • name: The identifier for your agent
  • bio: Array of strings describing your agent's personality and background
  • traits: Array of personality traits that define your agent's character
  • examples: Sample outputs that guide your agent's style
  • loop_delay: Time in seconds between agent actions

Platform Configuration​

The config section allows you to configure various platform integrations:

Twitter Configuration​

{
  "name": "twitter",
  "timeline_read_count": 10, // Number of tweets to read
  "own_tweet_replies_count": 2, // Max replies to own tweets
  "tweet_interval": 5400 // Time between tweets
}

Farcaster Configuration​

{
  "name": "farcaster",
  "timeline_read_count": 10,
  "cast_interval": 60
}

LLM Provider Configuration​

{
  "name": "openai",
  "model": "gpt-3.5-turbo"
}

Solana Configuration​

{
  "name": "solana",
  "rpc": "https://api.mainnet-beta.solana.com"
}

Ethereum Configuration​

{
  "name": "ethereum",
  "rpc": "https://eth.blockrazor.xyz"
}

Sonic Configuration​

{
  "name": "sonic",
  "network": "mainnet"
}

Time-based Activity Configuration​

Configure when and how your agent operates throughout the day:

{
  "use_time_based_weights": true,
  "time_based_multipliers": {
    "tweet_night_multiplier": 0.4,     // Reduces tweet frequency 1 AM - 5 AM
    "engagement_day_multiplier": 1.5    // Increases engagement 8 AM - 8 PM
  }
}

The agent will automatically:

  • Reduce tweeting during night hours (1 AM - 5 AM) to 40% of normal frequency
  • Increase engagement during day hours (8 AM - 8 PM) by 150%
  • Maintain normal activity at other times

This configuration allows you to:

  • Control posting frequency during different times of day
  • Adjust engagement levels based on time periods
  • Create more natural activity patterns

Fine-Tuning Your Model​

For more sophisticated outputs, you can fine-tune your own model. Follow the OpenAI fine-tuning guide for detailed instructions.

Best Practices​

  1. Detailed Bio: Provide a rich backstory and context for your agent
  2. Diverse Examples: Include varied examples that showcase different aspects of your agent's personality
  3. Balanced Weights: Adjust task weights to create natural behavior patterns
  4. Appropriate Intervals: Set reasonable delays between actions to avoid rate limits

Tips for Success​

  • Craft a unique and consistent personality for your agent
  • Use examples that demonstrate the desired tone and style
  • Monitor your agent's performance and adjust configuration as needed
  • Keep rate limits in mind when setting intervals