Hey Product Hunt 👋

I'm Jeremy, the maker of Product Pillars.

I kept hitting the same wall: I'd sketch a flow in a whiteboard tool, then re-describe the whole thing in text every time I wanted my AI agent (Claude Code, mostly) to understand it. The picture and the thing my tools could read were always two different artifacts, and they drifted constantly.

So I built the tool I wanted: a fast, focused canvas where the diagram and the Mermaid markdown are the same artifact. What you draw is byte for byte what you export, what you commit to a PR, and what you hand to an agent. Every node keeps a stable ID, so an agent can come back three turns later, say "update that node," and it still resolves to the right box.

What's live today:

• A canvas built for muscle memory: multi-select, undo/redo, copy-paste, keyboard-first node creation

• Purpose-built node types (start, user action, system action, decision, end)

• Clean Mermaid + Markdown export, byte-identical to the preview

• Public share links: send one URL, anyone opens it in a browser with no account, and there's a /md twin so agents can read it over a plain HTTP GET

• A hosted MCP server: generate a key in Settings, paste the config, and Claude Code reads and writes your diagrams directly. No link to pass, no export step.

Next up: co-editor invites and password-protected links.

It's free to start. Your first diagrams are on the house.

One thing I know you'll poke at: those stable IDs are UUIDs today, so the raw Mermaid is more verbose than the hand-written kind. That was a deliberate trade for agent addressing, but I'm not married to it. If readable slugs matter more to you, say so and I'll ship them.

I'd love your brutal feedback: what feels great, what feels missing, and whether "humans and agents on one diagram" resonates or sounds like overkill. I'm here all day. 🙏