A lightweight (not to say 'poor') AI agent implementation that can search the web and read website content.

This project was inspired by a homework assignment given during the excellent SolveIt developed by Answer.AI. It demonstrates how to build a functional agentic AI system using basic components:

• An LLM for reasoning and decision-making
• Simple function calling for tool use
• Web search and URL reading capabilities
• An iterative loop to complete multi-step tasks

The poor man's moniker reflects its minimalist approach, showing that effective AI agents don't require extensive code or complicated architectures.

• uv
• Task
• Optional: API key for OpenAI-compatible LLM provider (e.g. OpenRouter)
• Optional: API key for Jina AI to use ready-made tools

Make sure you meet the prerequisites

Install the package directly:

uv add git+https://github.com/slangenbach/poor-mans-agent.git

Or use it without installation:

uvx --from git+https://github.com/slangenbach/poor-mans-agent.git pma "Your prompt here"`

Install the package:

pip install git+https://github.com/slangenbach/poor-mans-agent.git

Copy .env.example to .env in the project root directory and set the following variables:

Once installed, you can run the agent from the command line:

pma "What does it take to build an AI agent in Python?"

The agent will:

1. Analyze your prompt
2. Decide which tools to use
3. Execute the necessary steps
4. Return a final answer

Example queries:

• pma "Find the latest news about AI agents"
• pma "Read the content from https://www.answer.ai/posts/2025-10-13-video-to-doc.html and summarize it"
• pma "Search for Python tutorials on building AI agents"

You can also use the agent in your Python code:

from poor_mans_agent.main import Agent
from poor_mans_agent.tools import search, read_url

agent = Agent(tools=[search, read_url])

response = agent.run("What does it take to build an AI agent in Python?")
print(response)

It's also possible to customize the agent:

agent = Agent(
    model="google/gemini-3-flash-preview",  # Use a different model
    tools=[search, read_url],
    system_prompt="You are an expert Gopher. Everything you do is written in Go.",  # Add custom system prompt
    max_iterations=5  # Allow more tool calls
)

• Missing API keys: Make sure all required environment variables are set in your .env file
• Invalid API keys: Verify your AI_MODEL_KEY and JINA_AI_KEY are valid and have sufficient credits
• Model not found: Check that AI_MODEL_NAME is a valid model identifier for your provider

The agent stops after 3 tool calls by default. If your task requires more steps:

agent = Agent(tools=[search, read_url], max_iterations=5)

If tool calls are timing out, increase the timeout in your .env file:

TOOL_CALL_TIMEOUT=60

Check out CONTRIBUTING.md for further information.