This is the first iteration of a local, distributed SWE agent system. The goal of this first iteration is to automate the creation of the frontend and backend of a basic website when given a user prompt and the figma assets.
Make sure to clone the repo with submodules
git clone --recurse-submodules https://github.com/acm-ucr/swe-agent.git
Create a conda enviorment with python 3.10
conda create -n myenv python=3.10
Install the requirements
pip install -r requirements.txt
In order to run the agent on a single device run the following
python run_agent.py \
--instruction_path /SWE-Agent/instructions/instruction.json \
--log_type test \
--main_device 1
In order to run the agent on the distributed setting run the following on the main device with your parameters
python run_agent.py \
--instruction_path /SWE-Agent/instructions/instruction.json \
--log_type run_name \
--main_device 1
Run the following on each coding agent device with your parameters
python run_agent.py \
--instruction_path /SWE-Agent/instructions/instruction.json \
--device_config /SWE-Agent/instructions/device_config.json
--distributed_config /SWE-Agent/instructions/distributed_config.json
--log_type run_name \
--main_device 0