TinyAgent

TinyAgent is a lightweight Python library for creating and managing conversational agents using a GPT-based model. The library provides an easy-to-use interface for building multi-turn conversations, defining system behavior, and interacting with a model API.

---

Table of Contents

1. Features
2. Installation
3. Usage
   • Initialize an Agent
   • Add Messages
   • Add Instructions and Examples
   • Configure Agent Behavior
   • Generate a Response
4. API Reference
   • Class: TinyAgent
5. License
6. Contributing
7. Contact

---

Features

• Customizable Agent Configuration: Set agent names, behaviors, and system messages.
• Conversation Management: Add system, user, and assistant messages to simulate multi-turn conversations.
• Instructions and Examples: Add explicit instructions and examples to guide model responses.
• Temperature and Token Management: Fine-tune the generation behavior using temperature and token limits.
• Flexible Output Formats: Support for text and JSON object responses.
• Lightweight and Intuitive API: Designed for rapid prototyping and development.

---

Installation

Clone the repository and install the required dependencies.

git clone https://github.com/Masum06/TinyAgent.git
cd TinyAgent
pip install -r requirements.txt

---

Usage

Initialize an Agent

from tinyagent import TinyAgent

# Create an agent with a name and system behavior
agent = TinyAgent(name="TinyAgent", system="You are a helpful assistant.")

Add Messages

# Add a system message
agent.add_system_message("Always provide concise and accurate answers.")

# Add a user message
agent.add_user_message("What is the capital of France?")

Add Instructions and Examples

# Add explicit instructions
agent.add_instruction("Respond politely and professionally.")

# Add input-output examples
agent.add_example("What is 2+2?", "4")

Configure Agent Behavior

# Adjust the temperature (creativity)
agent.set_temperature(0.7)

# Set a maximum token limit
agent.set_max_tokens(1500)

# Change the model
agent.set_model("gpt-4o")

Generate a Response

# Generate a text response
response = agent.call(prompt="Can you tell me about the Eiffel Tower?")
print(response)

# Generate a JSON response
json_response = agent.call_json(prompt="Provide details about Python programming.")
print(json_response)

---

API Reference

Class: TinyAgent

__init__(name="", system="")

• Parameters:
  • name (str): Name of the agent.
  • system (str): Initial system message to define the agent's behavior.

add_message(message_type, message)

• Description: Adds a message to the conversation.
• Parameters:
  • message_type (str): Role of the message (system, user, or assistant).
  • message (str): Message content.

add_system_message(message)

• Description: Adds a system message to the conversation.

add_user_message(message)

• Description: Adds a user message to the conversation.

add_instruction(instruction)

• Description: Adds an instruction to guide the agent's responses.

add_example(input, output)

• Description: Adds an input-output example to the conversation.

add_data(data)

• Description: Adds data as part of the user message.

set_temperature(temperature)

• Description: Adjusts the creativity of responses.

set_max_tokens(max_tokens)

• Description: Sets the maximum number of tokens for responses.

set_model(model)

• Description: Changes the underlying model used for responses.

call(prompt="", response_type="text")

• Description: Generates a response from the agent.
• Parameters:
  • prompt (str): User input or question.
  • response_type (str): Response format (text or json_object).
• Returns: Model-generated response (str or JSON object).

load_json(s)

• Description: Parses a string as JSON.
• Parameters:
  • s (str): Input string.
• Returns: Parsed JSON object or None.

call_json(prompt="")

• Description: Generates a JSON response.
• Parameters:
  • prompt (str): User input or question.
• Returns: Parsed JSON response.

---

License

This project is licensed under the MIT License. See the LICENSE file for details.

---

Contributing

Contributions are welcome! Please submit a pull request or open an issue for any bugs or feature requests.

---

Contact

For questions or support, feel free to reach out at m.hasan@rochester.edu.