HypothesizerAgent Documentation

HypothesizerAgent is a multi-agent system that iteratively refines solutions to complex problems through a structured debate process. It employs three specialized agents that work together to generate, critique, and provide alternative perspectives on potential solutions.

Basic Usage

from ursa.agents import HypothesizerAgent

# Initialize the agent
agent = HypothesizerAgent()

# Run the agent with a question
solution = agent.invoke(
    prompt="Find a city with at least 10 vowels in its name.",
    max_iter=3
)

# Print the final solution
print(solution)

Parameters

Initialization

Parameter	Type	Default	Description
llm	str or BaseChatModel	"openai/o3-mini"	The language model to use for all agents
**kwargs	dict	{}	Additional parameters passed to the base agent

Run Method

Parameter	Type	Default	Description
prompt	str	Required	The question or problem to solve
max_iter	int	3	Maximum number of refinement iterations
recursion_limit	int	99999	Maximum recursion depth for the graph

How It Works

The system uses a three-agent debate process:

1. Agent 1 (Hypothesizer): Generates initial solutions and refines them based on feedback
2. Agent 2 (Critic): Identifies flaws, assumptions, and areas for improvement
3. Agent 3 (Competitor/Stakeholder): Provides alternative perspectives from different stakeholders

The process iterates through these agents multiple times, with each iteration building on the feedback from previous rounds.

Features

• Web Search Integration: Uses DuckDuckGo to gather information for each agent
• Iterative Refinement: Solutions improve through multiple rounds of critique and revision
• LaTeX Report Generation: Creates a comprehensive LaTeX document summarizing the process
• URL Tracking: Records all websites visited during the research process

Output

The agent produces:

1. A final refined solution (returned by the run method)
2. A LaTeX document with:
3. Executive summary of the iterative process
4. Final solution in full
5. Detailed appendix of all iterations
6. List of websites visited during research
7. A text file containing the full history of all iterations

Example

from ursa.agents import HypothesizerAgent

# Initialize with a specific LLM
agent = HypothesizerAgent(llm="openai/gpt-4o")

# Run with 5 iterations
result = agent.invoke(
    prompt="What strategies could a small local bookstore use to compete with large online retailers?",
    max_iter=5
)

print(result)

Notes

• Each iteration includes a solution, critique, and competitor perspective
• The agent performs web searches to gather information at each step
• The final LaTeX document provides a comprehensive record of the reasoning process
• Higher max_iter values produce more refined solutions but take longer to complete