Agents in Underdog Cowboy

Overview

Agents in the Underdog Cowboy system are intelligent entities designed to perform specific tasks, engage in dialogues, and assist users in various operations. These agents serve as intermediaries between users and Language Models (LLMs), encapsulating conversation context, specialized knowledge, and specific capabilities.

Types of Agents

1. User-Defined Agents

• Created through interactions in the Timeline Editor

• Stored as JSON files in the user's home directory (~/.underdogcowboy/agents)

• Reflect personalized conversation patterns and knowledge

2. Specialized Agents

• Pre-packaged with the Underdog Cowboy release

• Installed via pip along with the main package

• Designed for specific, advanced functionalities

3. Regular Dialogs

• Stored in the same JSON format as agents

• May not have specific agent behaviors

• Useful for storing and reusing conversation patterns

Agent Structure

Agents are represented by the Agent class, which includes:

• Unique identifier (ID)

• Name

• Associated filename and package

• Loaded content from JSON file

• Dialog manager for handling interactions

Creating Agents

1. Using Timeline Editor

   • Launch with timelineeditor command

   • Engage in dialogue with an LLM

   • Save the interaction as an agent

2. Manual JSON Creation

   • Create a JSON file with required agent structure

   • Place in appropriate directory for loading

Using Agents in Scripts

from underdogcowboy import adm, your_agent_name

# Enable dialogue
adm | [your_agent_name]

# Send a message
response = your_agent_name >> "Your message here"
print(response)

Interactive Development

Agents can be refined interactively:

1. Add +adm in your script to enter interactive mode

2. Engage in dialogue to test and improve agent responses

3. Use resume to continue script execution with updated behavior

Advanced Features

• Operator Overloading: Use >> for sending messages

• Dialog Manager Integration: Agents work closely with the Dialog Manager for message handling

• Dynamic Loading: Agents are loaded dynamically, allowing for easy expansion and customization

Best Practices

1. Start with a clear purpose for your agent

2. Use the Timeline Editor for initial creation and testing

3. Refine agents through interactive development in scripts

4. Regularly update and maintain your agents based on performance

Conclusion

Agents in Underdog Cowboy provide a flexible and powerful way to interact with LLMs, enabling personalized, context-aware, and task-specific AI assistance. Whether using pre-built specialized agents or creating your own, the system offers a range of tools for developing and deploying intelligent agents in your projects.