Skip to content

Latest commit

 

History

History
168 lines (116 loc) · 4.91 KB

UsingAgentForge.md

File metadata and controls

168 lines (116 loc) · 4.91 KB

Using AgentForge

This guide covers the basics of running single and multi-agent scripts in AgentForge, focusing first on defining agents via prompt files, then introducing the option to create custom classes if you need more specialized behavior.

Running an Agent

1. Create the Prompt Template

Inside your project's .agentforge/prompts/ directory, place a YAML file named EchoAgent.yaml:

prompts:
  system: You are an assistant that echoes the user's input.
  user: {user_input}

Here, {user_input} is a placeholder that will be filled at runtime. For more details on how placeholders and agent prompts work, see the Agent Prompts guide.

2. Write a Script to Run the Agent Interactively

In your project root, create a Python script (e.g., run_agent.py):

from agentforge.agent import Agent

def main():
    # Instantiate the EchoAgent using its YAML prompt file
    echo_agent = Agent('EchoAgent')

    print("Welcome to the EchoAgent! Type 'quit' to exit.")
    while True:
        user_input = input("You: ")
        if user_input.lower() == 'quit':
            print("Goodbye!")
            break

        response = echo_agent.run(user_input=user_input)
        print("EchoAgent:", response)

if __name__ == "__main__":
    main()

3. Run Your Script

Activate your virtual environment if you have one, then run:

python run_agent.py

An example interaction might look like this:

Welcome to the EchoAgent! Type 'quit' to exit.
You: Hello there!
EchoAgent: Hello there!
You: This is a test.
EchoAgent: This is a test.
You: quit
Goodbye!

Running a Multi-Agent Script

When you want multiple agents collaborating in a single script, each agent can be defined by its own YAML file and then instantiated the same way.

1. Create Prompt Files

Place these files in .agentforge/prompts/:

QuestionGeneratorAgent.yaml

prompts:
  system: You are a helpful assistant that generates insightful questions based on a given topic.
  user: |
    Topic: {topic}

AnswerAgent.yaml

prompts:
  system: You are a knowledgeable assistant that provides detailed answers to questions.
  user: |
    Question: {question}

2. Write the Multi-Agent Script

In the root of your project, create run_topic_qanda.py:

from agentforge.agent import Agent

def main():
    # Instantiate agents using their YAML prompt files
    question_agent = Agent('QuestionGeneratorAgent')
    answer_agent = Agent('AnswerAgent')

    topic = "artificial intelligence"
    question = question_agent.run(topic=topic)
    print(f"Generated Question: {question}")

    answer = answer_agent.run(question=question)
    print(f"Answer: {answer}")

if __name__ == "__main__":
    main()

3. Execute the Script

python run_topic_qanda.py

The output will vary based on your chosen LLM and configuration. For example, you might see:

Generated Question: What are the ethical implications of artificial intelligence in modern society?
Answer: The ethical implications of artificial intelligence (AI) in modern society include concerns about privacy...

Creating Custom Agent Classes (Advanced)

In many cases, simply referencing a YAML file is sufficient for defining agent behavior. However, if you need extra control—such as adding custom methods or overriding certain behaviors—you can create a custom agent class. Below is a quick look at how you might do it:

from agentforge.agent import Agent

class CustomAgent(Agent):
    def process_data(self):
        # Perform any custom pre-processing before sending data to the LLM
        pass

    def build_output(self):
        # Adjust or filter the LLM’s response
        return super().build_output()

# Instantiate and use it just like any other agent
my_custom_agent = CustomAgent('EchoAgent')
reply = my_custom_agent.run(user_input="Hello!")
print(reply)

This approach gives you a straightforward way to inject logic before or after the primary LLM call, making it easy to tailor your agents for specialized tasks.

Additional Resources

You can explore more complex agent features, define defaults in .agentforge/settings/ for your model settings, or integrate with different LLM APIs by checking out the Settings and LLMs guides. To learn more about advanced customization or how to manage large-scale agent projects, visit the Agents Documentation.

If you run into any issues, refer to the Troubleshooting Guide for common fixes and workarounds.

Next Steps: