MAICC: Multi-agent In-context Coordination via Decentralized Memory Retrieval

This is the implementation of the AAAI 2026 paper "Multi-agent In-context Coordination via Decentralized Memory Retrieval". The appendix of the paper is available at this MAICC_supp.pdf.

Install Python environment

Install Python environment with conda:

conda create -n MAICC python=3.8 -y
conda activate MAICC 
pip install pip==24.0
pip install setuptools==63.2.0
pip install wheel==0.38.4
pip install -r requirements.txt
pip install -e lb-foraging/
pip install -e smacv2/

Install StarCraft

bash install_sc2.sh

Run experiments

We provide the data link to the full dataset by this Google Drive URL. Unzip datasets.zip into the MAICC/datasets folder.

You can run the following scripts to perform experiments in the paper.

bash lbf_small_train.sh
bash lbf_train.sh
bash sc2v2_train.sh