Team
AG-WebGL
Project Concept
Type a process at the chat box and a five-agent pipeline streams scene mutations into a live WebGL canvas via the AG-UI / CopilotKit event stream.
Entry
Status: Submitted
Last saved: May 09 at 5:19 PM PDT
Team Roster
Message board not available for this team yet.
Eduardo Team Lead RSVP Approved
Software Engineer at Freelancer
I built the entire project end to end as a solo submission.
I designed and implemented the multi-agent LangGraph pipeline running on FastAPI, including the Director, Layout, Asset, Animation, and Lighting agents. Each agent communicates through a shared in-memory Scene Store using a lazy context hydration pattern, pulling only the object detail needed per run rather than serializing the full scene on every call.
I built the AG-UI custom event vocabulary that defines how agents mutate a live Three.js canvas, and wired the full CopilotKit integration on the frontend, using useCopilotAction to route each streamed event to a purpose-built SceneController that handles all Three.js mutations. I also designed the hybrid scene serializer that feeds agents a rolling context window, a full object index, and a spatial zone map so agents can reason about the canvas they are editing.
I built the Three.js canvas, animation loop, and object factory from scratch with a strict userData tagging convention that keeps every object readable by the agent pipeline without ever touching raw geometry buffers.
The full stack spans AG-UI for agent-to-frontend event streaming, CopilotKit for the React integration layer, A2UI for declarative component conformance, LangGraph and LangChain for agent orchestration, Three.js for the live 3D canvas, Next.js for the frontend, and FastAPI for the agent backend.
Agnostic Developer currently exploring AI architecture and tools.
AI Harnesses, Ag-ui, skills
Baton - agent to agent context sharing