Jazzboard

Jazzboard is a multiplayer whiteboard where you and your AI agents create together live. Ask Codex to map your real codebase, sketch an idea, or draw something ridiculous, then watch it come to life.

Comment 1

Inspiration

My remote team and I miss the energy of being in the same room during a feature meeting, gathered around one whiteboard and building an idea together. Video calls and shared documents never quite recreated that experience. We built Jazzboard to bring it back, while adding a new kind of teammate to the room: AI agents.

What it does

Jazzboard is a multiplayer whiteboard where people and AI agents create together live. You can ask Codex to map your real codebase, design an architecture, explore an idea, or draw something fun. The agent works directly on the shared canvas while everyone watches, collaborates, and edits alongside it.

How we built it

Jazzboard uses a first-party semantic canvas, real-time multiplayer synchronization, Redis, Vercel, and browser-native WebMCP tools. Every object has meaningful identity, geometry, relationships, revisions, and attribution.

Agents can create entire diagrams in efficient transactions, visually trace their work onto the board, inspect the result, detect readability problems, and safely correct them. We also built a research harness that measures speed, accuracy, and visual quality using isolated agent runs and blinded reviews.

Challenges we ran into

The hardest challenge was making AI work feel both fast and alive. Agents need to create complex artifacts efficiently, but people should still be able to watch that work unfold naturally.

We also had to solve real-time conflicts between humans and agents, prevent delayed updates from snapping objects backward, preserve connector relationships, and give agents enough visual context to recognize cramped or unreadable results without making creative decisions for them.

Accomplishments that we're proud of

We built a canvas that agents can truly understand and operate, not just click through with browser automation.

Agents can build editable diagrams from real project context, draw original artwork, inspect their own work, identify visual problems, and self-correct. Humans can collaborate on the same board in real time while animated, uniquely generated bots visibly move around and construct their work.

We are especially proud that Jazzboard preserves agent autonomy. It provides evidence, context, and precise tools, while leaving composition and creative decisions to the agent.

What we learned

An agent is only as capable as the environment and context we give it. A powerful model can still create a poor diagram if it cannot understand object relationships, spatial conflicts, or what its finished work actually looks like.

We learned that the best agent interfaces do more than expose CRUD operations. They provide semantic state, actionable errors, visual inspection, revision safety, and clear recovery paths. Good harness engineering can improve both quality and speed without limiting what the agent is allowed to create.

What's next for Jazzboard

We want Jazzboard to become the best canvas in the world for agent-created architecture diagrams and visual art.

Next, we are expanding our research benchmarks, improving creation speed, supporting increasingly complex boards, and giving agents even stronger tools for visual reasoning and self-correction. Our larger vision is a universal creative space where people and AI can think visually together, no matter where they are.

Built With

Share this project:

Updates

Submission history