Coming soon
Give your agents a whiteboard.
pks-agent-whiteboard
A whiteboard that lives inside your editor, where your coding agent pins screenshots, sticky notes and architecture diagrams next to your code — so it can finally show you what it's thinking, not just tell you.
Your agent can write a thousand lines. It still can't point at anything.
Coding agents are brilliant in a single column of scrolling text — and then they hit a wall. They can describe a layout, narrate an architecture, claim a screen renders correctly. But a wall of prose is the worst possible medium for a diagram, a mockup, or "look, it works." You end up reconstructing the picture in your own head, every time, while the real understanding stays trapped in the chat log.
Agents talk in one scrolling column — diagrams, mockups and screenshots have nowhere to live.
"It renders fine" is a claim, not a picture. You can't trust what you can't see.
A real workshop needs a wall to think on. Your terminal doesn't have one — and flying someone in to draw on a real one is slow and expensive.
How it works
Open the Whiteboard
Open the Whiteboard from the VS Code activity bar. The canvas is just there, beside your code — nothing to host, nothing to log into.
Paste one line
Click "Copy MCP Install Command" in the panel header and paste one line in your terminal. That's the whole setup.
Keep vibecoding
Keep vibecoding as normal. Your agent now has the tools to pin images, sticky notes and diagrams — it places them for you, automatically.
Arrange the wall
Pan, zoom and drag to arrange the wall. Click a diagram to edit it, click an image to open it full-size. The room is yours.
Pinned screenshots
Your agent posts a real image — a render, a chart, a UI mock — straight onto the canvas. Click to open it in your editor; shift-click for a lightbox.
Sticky-note mindmaps
Editable sticky notes in six colors let your agent lay out what it understands about the task — a mindmap you can read at a glance and edit yourself.
Live drawio diagrams
Architecture sketches drop in as real, editable drawio diagrams. Pick one up, open the full editor, and keep building from where the agent left off.
Pan, zoom, arrange
A true 2D canvas: drag empty space to pan, Cmd-scroll to zoom, drag a card to move it, drag a corner to resize. Lay the room out the way you think.
One wall, many agents
Run several agent sessions in parallel and they all push to the same canvas. The whole team's thinking lands on one shared wall.
Local and private by default
The canvas binds a random loopback port and writes a per-session bearer token to a `0600` file. It stays on your machine — nothing leaves your editor.
Built from the foundation up. The IPC contract, the MCP tool names and the canvas itself are ours — no third-party whiteboard SaaS, no account to create, no platform that can break or breach underneath you.
Loopback-only, token-gated, local by default. The data never leaves your machine, so the security model is small enough to reason about end to end.
Speaks plain MCP, so it runs everywhere. Claude Code today, anything that speaks the protocol tomorrow — the whiteboard doesn't care which agent is drawing on it.
A surface, not a silo. It's one owned layer in a suite that supports itself top to bottom — you run only the pieces you need, and they compose.
Composes with
The room where your agents think out loud.
A pan-and-zoom canvas right in VS Code. Your agent drops images, sticky notes and live diagrams while you vibecode in the terminal beside it. Run an agentic workshop that pulls the insight out of the room — no consultants on a plane.