Home

Awesome

CoAgents

The code for "cooperative and interactive agents framework"

We propose the ConAgents, a Cooperative and interactive agents framework . The grounding agent generates tool-use instructions while execution agent executes selected tools. The observing agent extracts relevant values from execution results. When receiving feedback from the tool environment, the execution agent and observing agent adapt themselves iteratively.

image-20240324212748206

image-20240324212928539

Quick start

langchain==0.0.338 # new version is also allowed
colorama==0.4.6
tiktoken==0.5.1
openai==1.7.1
python run_tmdb.py \
--model_name 'gpt-3.5-turbo' \
--log_file './log_file.json'  \
--data_file <your dataset file> \
--access_token_file <the file containing your access token of the TMDB platform>

Todo

We will add more details as soon as possible.

citation

@article{shi2024learning,
  title={Learning to Use Tools via Cooperative and Interactive Agents},
  author={Shi, Zhengliang and Gao, Shen and Chen, Xiuyi and Yan, Lingyong and Shi, Haibo and Yin, Dawei and Chen, Zhumin and Ren, Pengjie and Verberne, Suzan and Ren, Zhaochun},
  journal={arXiv preprint arXiv:2403.03031},
  year={2024}
}