Argonaut dashboard turns multi-agent optimization black box into interactive visual tool

Distributed optimization is notoriously opaque: when thousands of agents make interdependent choices, the path to a global solution becomes a black box. Existing tools visualize only the final result or lock algorithms to fixed datasets, leaving the search process invisible. To break that open, researchers introduced Argonaut—a lightweight, containerized dashboard that makes every step of multi-agent discrete-choice optimization explorable. With a browser-based interface and extensible Java and Python backends, users can upload real datasets, construct agents and options, and modify the decision space in real time while running multiple optimization algorithms side by side.

Argonaut unifies system construction, optimization, and analysis in a single human-in-the-loop loop—the first tool of its kind. Its evaluation on household-electricity, shared-mobility, and sensor-data-exchange datasets shows serious scale: it handles up to 5,600 agents and brute-force search spaces reaching 1 million solutions. Performance stays practical, with a typical run of 200 agents over 100 decision attributes completing in under 30 seconds. Presented at ACSOS 2026 in the artifacts track, Argonaut turns distributed optimization from a one-shot computation into an inspectable, interactive process—giving researchers and engineers a tangible way to see how local decisions shape global outcomes.