ProAgent
ProAgent is an LLM-based agent system that automates complex workflows by constructing and executing them from human instructions, coordinating specialized agents. It addresses the limitations of traditional Robotic Process Automation (RPA) in tasks requiring human-like intelligence. It is intended for developers and researchers exploring agentic process automation.
✨ Key features
- Constructs workflows from natural language instructions
- Coordinates specialized agents for dynamic decision-making
- Supports development, refine, and production running modes
- Includes a readable record system for replaying runs
- Can ask humans for help via function-calling
- Integrates with self-hosted n8n for real-world app connections
🎯 Use cases
- Automating multi-step business processes that require adaptive decision-making
- Building and testing workflows in a development environment
- Refining existing workflows with new requirements
- Reproducing and analyzing previous ProAgent runs
📦 Installation
🧰 Requirements: Requires Python, npm, an OpenAI API key (for development mode), and optionally a self-hosted n8n instance.
pip install -r requirements.txt
Install n8n (for development mode):
npm install n8n -g
Set webhook URL and start n8n:
export WEBHOOK_URL=https://n8n.x-agent.net/redirect/http%3A%2F%2Flocalhost%3A5678/
n8n
Export credentials and workflows:
n8n export:credentials --all --decrypted --output=./ProAgent/n8n_tester/credentials/c.json
n8n export:workflow --all --output=./ProAgent/n8n_tester/credentials/w.json
Move the files to ./ProAgent/n8n_tester/credentials/.
🚀 Usage
Set environment variables for OpenAI (if using development mode):
export OPENAI_API_KEY=your_key
export OPENAI_API_BASE=your_base
Run ProAgent:
python main.py
Configure mode in ProAgent/config.py: development, refine, or production.
⚠️ Good to know
The n8n compiler may not be compatible with newer versions of n8n, and self-hosting n8n can be challenging, especially with network restrictions.
❓ FAQ
What are the running modes?
Development mode is for building workflows from scratch, refine mode loads an existing workflow and refines it with new requests, and production mode loads an existing workflow to reproduce a run.
Do I need n8n to run ProAgent?
n8n is required for development mode, but you can use the provided case records to reproduce paper results without n8n.
Which OpenAI model does ProAgent use?
It is based on GPT4-0613, but newer models like GPT4-1106-preview may improve performance.
How can I load a previous run?
All runs are recorded in the ./records directory, and you can load them in refine or production mode.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.