From water wheels to Robotic Process Automation (RPA), automation technology has evolved throughout history to liberate human beings from arduous tasks. Yet, RPA struggles with tasks needing human-like intelligence, especially in elaborate design of workflow construction and dynamic decision-making in workflow execution. As Large Language Models (LLMs) have emerged human-like intelligence, this paper introduces Agentic Process Automation
(APA), a groundbreaking automation paradigm using LLM-based agents for advanced automation by offloading the human labor to agents associated with construction and execution. We then instantiate ProAgent
, an LLM-based agent designed to craft workflows from human instructions and make intricate decisions by coordinating specialized agents. Empirical experiments are conducted to detail its construction and execution procedure of workflow, showcasing the feasibility of APA, unveiling the possibility of a new paradigm of automation driven by agents
And this is the official code of Agentic Process Automation
paper, you can temporally download our paper here
pip install -r requirements.txt
especially, We use the OpenAI version before Dev Day, so you can't use the latest version of OpenAI
Our projects use a self-host n8n, you can either prepare a n8n environment and connect ProAgent with a realworld APP service.
Or you can load our record to re-produce the case reported in our paper without n8n environment
prepare a n8n service is not a eazy thing, you may face some problems or bug that we haven't encountered before, and you must handle the situation. For example, you may find some GFW problems in China when connecting your APP accounts with n8n
Our projects use a self-host n8n, you must first install a n8n following the guide . You can use this command in linux/macOS
npm install n8n -g
self-host n8n doesn't support https
service. However, we have already built a redirect-service, you can temporally use our service(may not be very stable, and we will opensource the redirect-service code)
export WEBHOOK_URL=https://n8n.x-agent.net/redirect/http%3A%2F%2Flocalhost%3A5678/
n8n
You need to regist or connect a existing APP with n8n before launching ProAgent
. Connecting an APP may have some APP-specific operations, you can follow the n8n credential guide here
Our code base needs to load the workflow ID and credential ID. So you must make some workflow and regist some apps before, then do the following commands to decode the credentials from n8n service
n8n export:credentials --all --decrypted --output=./ProAgent/n8n_tester/credentials/c.json
move c.json
to ./ProAgent/n8n_tester/credentials/c.json
n8n export:workflow --all --output=./ProAgent/n8n_tester/credentials/w.json
move w.json
to ./ProAgent/n8n_tester/credentials/w.json
The running depends on a config
๏ผ which is in ProAgent/config.py
, you can set the running environment:
- development: This is the mode reported in paper, .
- refine: load from a existing workflow, and then refine the workflow with some new request
- production: load from a existing workflow, you can use this mode to re-produce a existing run of
ProAgent
we have provide the case reported in our paper in ./apa_case
, you can use production
mode to load the run directly
we disable the test-on-change feature in production mode, And the APA-code will be test only once in the end of the run
In the opposite, the refine mode enable test-on-change feature
use the following command to start ProAgent
, if you use development
mode, remember to start n8n before
python main.py
Note that we have wrote a readable record system. All of the
ProAgent
runs will generate a new record in./records
, so you can load the run from record inrefine
orproduction
mode.
If you use the development mode, you must prepare OpenAI key first. Set the following vars in your envirnoment
OPENAI_API_KEY, OPENAI_API_BASE
The method in our code is almost the same as descripted in our paper. However something have changed after the paper released:
-
HCI: we observed some problems when
ProAgent
has some misunderstood of the problem, so we developed a feature to enable ProAgent to ask human for help by a new function-call (like XAgent). This proactive manner encourageProAgent
to build and test workflows together with human. -
n8n-feature: n8n have rapidly released some new feature in parallel of our work. Our code is basically built upon an older version of n8n and we have wrote an
n8n-compiler
in our code base. So that compiler may not be compatible with the newer verison of n8n -
OpenAI:
ProAgent
is based onGPT4-0613
. However OpenAI have releasedGPT4-1106-preview
in Dev Day, which has a 38% improvment in function-calling. So you may findProAgent
better than our paper thanks to OpenAI's update~