AgentCommander

Automating the loop of discovery.

AgentCommander was born from the actual demands of scientific research.

Refined through rigorous practical application, it is a graph-based workflow engine designed to orchestrate AI Agents for complex, iterative tasks. Built to leverage the diverse ecosystem of LLM CLIs (Gemini, Qwen, Claude, etc.), it enables Machine Learning engineers to construct highly customizable, infinite-loop workflows.

Philosophy: Human-Centric Evolution

Unlike "set and forget" evolutionary algorithms (like AlphaEvolve or OpenEvolve) where agents are left to mutate randomly in a vacuum, AgentCommander is built for Collaborative Discovery.

• You are the Commander: You define the search space, the hypothesis generation logic, and the evaluation metrics via a visual workflow.
• AI is the Executor: The agent handles the exhaustive, repetitive loop of coding, debugging, and refining.
• Transparent & Controllable: We prioritize white-box transparency. Every step is logged, every file is accessible, and you can intervene at any moment.

Key Capabilities

• Symbolic Regression: Automating the search for mathematical expressions.
• ML Structure/Hyperparameter Optimization: Intelligent tuning without manual intervention.
• Autonomous Model Refinement: Self-correcting loops for model improvement.

Contact

📧 Email: miaoxin.liu@u.nus.edu