EvoGit wins the multi-agent track at AgentX

Our team at PolyU won first place in the multi-agent track at Berkeley’s AgentX competition. I worked on EvoGit with PhD student Beichen Huang and my colleague Kay Chen Tan. Competition results

In EvoGit, a group of coding agents works on a shared Git repository. Each agent can pick up another agent’s commit, try a change, or combine ideas from two branches. The commit history lets us trace how a piece of code developed and see which agent contributed each change.

EvoGit agents building on and combining Git commits

We tried EvoGit on two tasks. In one, the agents built a web application from an empty repository. In the other, they wrote a program that could itself develop and test solutions to a bin-packing problem. EvoGit paper

Enter to select · ↑ ↓ to navigate · Esc to close