Diagrams You Can Talk To: Using Excalidraw with OpenClaw

Excalidraw is the whiteboard everyone reaches for when an idea needs a picture — system architecture, a quick flowchart, a wireframe, the boxes-and-arrows version of a plan. Its charm is that it gets out of the way: hand-drawn look, no chrome, no learning curve.

But there is still a gap between having the idea and having the diagram. You picture the flow in your head, then spend ten minutes dragging rectangles, aligning them, drawing the arrows, nudging labels. The thinking took seconds; the drawing takes the rest of the meeting.

OpenClaw closes that gap, and it can do so because of one underrated fact about Excalidraw: its drawings are not opaque images. They are plain, structured files.

Why Excalidraw is unusually easy for an assistant to work with

An .excalidraw file is just JSON — a list of elements with types, positions, sizes, text, and the arrows that bind them. Nothing is locked inside a binary blob. That means an assistant can do far more than generate a picture; it can read the actual structure of a diagram, reason about it, and write a valid one back.

That distinction matters. A model that only produces images gives you a flat PNG you then have to redraw by hand to change anything. An assistant that writes the underlying Excalidraw file gives you a diagram you can keep editing — in Excalidraw, by hand, or by asking for the next change. The drawing stays live.

What the use case actually looks like

Here is the shape of it, in the messaging app you already use:

• "Draw the request flow: browser → Cloudflare Worker → the contact API → Resend, with a note that input is HTML-escaped." — and a flowchart comes back, boxes labelled, arrows in the right direction.
• "Take the architecture diagram in my vault and add a Redis cache between the API and Postgres." — it reads the existing file and inserts the node into the flow rather than starting over.
• "Turn these meeting notes into a simple swimlane of who does what." — prose in, a structured diagram out.
• "Make a wireframe of a three-section landing page: hero, features, contact." — a rough layout you can hand to a designer or build against.

You are no longer drawing. You are describing, reviewing, and refining — which is the part of diagramming that was ever actually yours to do.

How the pieces fit together

The drawing itself is the artifact; the value is in where it lives and how it stays current.

Most people who lean on Excalidraw keep their drawings alongside their notes, and the Obsidian skill is what makes that loop close: the Excalidraw plugin for Obsidian stores each drawing as a file in your vault, so an assistant with vault access can open the diagram referenced in a note, update it, and write it back next to the text that describes it. The picture and the prose stay in sync because they live in the same place.

For diagrams that belong to a codebase — an architecture doc, a sequence diagram in a docs/ folder — the GitHub skill means the assistant can commit the .excalidraw file like any other source. Because the format is text-based JSON, it diffs and versions cleanly: you can see in a pull request that a node was added, the same way you review a code change. A diagram in git is one that actually reflects the system, because it changes when the system does.

Where this earns its keep

The reason this is worth setting up is not novelty — it is that the diagrams which matter most are exactly the ones that rot. The architecture sketch made on day one is accurate for a week and misleading by month three, because keeping it current meant reopening the whiteboard and dragging boxes, and nobody had ten spare minutes. When updating a diagram costs one sentence, it gets updated. The picture stays true.

It is the same leverage we keep coming back to: an assistant absorbing the mechanical part of a task so the judgement part is all that is left to you. We wrote about a version of this in your assistant's daily standup — a voice note turning into an organised day. A diagram that redraws itself from a sentence is the same idea aimed at the whiteboard.

Getting started

If you already run OpenClaw, the path is short: give the assistant access to wherever your drawings live — your Obsidian vault, a repo, a shared folder — and start describing diagrams instead of drawing them. The work worth doing carefully is the setup around it: which folders it can write to, how generated diagrams get reviewed before they land, and naming conventions so the right diagram updates rather than a duplicate appearing.

Getting that wired up so it is genuinely useful rather than a novelty is the kind of thing we do with clients. If "describe the diagram, get an editable diagram back" sounds like how you would rather work, tell us what you are trying to map out.

Related posts

• Your Assistant's Daily Standup: One Voice Note, a Whole Day Organised
• Run Your Smart Home from WhatsApp: OpenClaw + Home Assistant