ProAgent introduces Agentic Process Automation (APA), an LLM-based approach to overcome the limitations of traditional Robotic Process Automation (RPA) for tasks requiring human-like intelligence. It targets developers and researchers seeking to automate complex workflows by leveraging LLM agents for intelligent workflow construction and dynamic execution.

How It Works

ProAgent utilizes LLM-based agents to interpret human instructions, construct detailed workflows, and coordinate specialized agents for execution. This approach offloads complex decision-making and workflow design to the LLM, enabling more sophisticated and adaptable automation compared to rule-based RPA systems.

Quick Start & Requirements

• Install packages: pip install -r requirements.txt
• Requires a self-hosted n8n instance.
• OpenAI API key (specifically, a version before OpenAI's Dev Day is noted as used in the paper, though newer versions might offer improvements).
• Credentials and workflows must be exported from n8n:
  • n8n export:credentials --all --decrypted --output=./ProAgent/n8n_tester/credentials/c.json
  • n8n export:workflow --all --output=./ProAgent/n8n_tester/credentials/w.json
• Official documentation: https://github.com/OpenBMB/ProAgent

Highlighted Details

• Enables human-in-the-loop interaction for clarifying misunderstandings via function calls.
• Includes an n8n-compiler due to reliance on an older n8n version.
• Potential for improved performance with newer OpenAI models (e.g., GPT-4-1106-preview).
• Provides case studies in ./apa_case for reproduction.

Maintenance & Community

• The project is associated with the paper "ProAgent: From Robotic Process Automation to Agentic Process Automation" (arXiv:2311.10751).
• A redirect service is temporarily provided, with plans to open-source its code.

Licensing & Compatibility

• The repository does not explicitly state a license in the provided README.

Limitations & Caveats

• Setting up a self-hosted n8n instance can be challenging and may encounter issues like GFW problems in China.
• The included n8n-compiler might not be compatible with newer n8n versions.
• The project's code is based on an older OpenAI version, though newer versions may offer benefits.