AgentForge-Autogen

# Agent Creator: Multi-Agent AutoGen Project

Agent Creator is a multi-agent AutoGen project where agents interact with each other to generate, refine, and evolve startup ideas.

At the center is **Agent Creator**: an agent that builds and launches new agents using AutoGen, unlocking endless AI possibilities.

## What This Project Does

- Uses a creator agent (`Creator`) to generate new agent files from a template.
- Dynamically registers each generated agent at runtime.
- Lets agents bounce ideas between one another for refinement.
- Saves each final output into `ideaN.md` files.
- Provides a simple Gradio UI to choose how many agents to run.

## Architecture Overview

- `creator.py`: Creator agent that asks an LLM to produce new `Agent` variants from `agent.py`.
- `agent.py`: Base template used to generate new agents with new personalities/system prompts.
- `agent1.py`, `agent2.py`, `agent3.py`: Existing generated examples.
- `messages.py`: Shared message model and helper to select another agent for idea refinement.
- `world.py`: Runtime host/orchestrator that spawns workers and collects ideas.
- `ui.py`: Gradio interface for interactive runs.

## Tech Stack

- Python 3.12+
- AutoGen (`autogen-agentchat`, `autogen-ext`, `autogen-core`)
- Groq OpenAI-compatible API (LLM backend)
- Gradio (web UI)

## Prerequisites

1. Python 3.12 or later
2. A valid Groq API key

Create a `.env` file in the project root:

```env
GROQ_API_KEY=your_groq_api_key_here
```

## Installation

### Option 1: Using uv (recommended)

```bash
uv sync
```

### Option 2: Using pip

```bash
python -m venv .venv
# Windows PowerShell
. .venv/Scripts/Activate.ps1
pip install -r requirements.txt
```

## Run the Project

### Run via UI

```bash
python ui.py
```

Then open the Gradio URL shown in the terminal.

### Run via orchestrator only

```bash
python world.py
```

This will generate `idea1.md`, `idea2.md`, etc. based on `HOW_MANY_AGENTS` in `world.py`.

## How Agent Creation Works

1. `world.py` sends a target filename like `agent4.py` to `Creator`.
2. `Creator` reads `agent.py` as a template.
3. The LLM returns updated Python code with a new `system_message`.
4. `Creator` saves the file, imports it, and registers it live.
5. The new agent is messaged to produce an idea, which can be refined by another random agent.

## GitHub Push Notes

The following notebooks are intentionally excluded from Git:

- `1_lab1_autogen_agentchat.ipynb`
- `2_lab2_autogen_agentchat.ipynb`
- `3_lab3_autogen_core.ipynb`
- `4_lab4_autogen_distributed.ipynb`

This is configured in `.gitignore`.

## Suggested Repo Structure

```text
agent.py
agent1.py
agent2.py
agent3.py
creator.py
messages.py
world.py
ui.py
pyproject.toml
requirements.txt
README.md
.gitignore
sandbox/
```