← All projects

Noodla

Me playing with how far a canvas can go, and what thinking, iterating and creating look like when everything else jumps straight to the end. Built on tldraw, with a 3D mascot that votes, a crowd that follows the best ideas and boards that finish with a conclusion.

A Noodla board of six notes about a Saturday workshop, with a crowd of small 3D characters gathered round the note stamped Decided
Press C and the vote crowd turns up. They follow votes and Decided stamps · sample board

The short version

pull requests in two weeks
44
browser tests on every release
79
running cost, on free tiers
£0

What it is

I work at Miro, we've just been bought by Bending Spoons, and I've spent a big chunk of my career in canvases. I worked on InVision Freehand and absolutely loved it (I used to tell InVision it was a bit of a sleeper tool inside our org). Then Miro, which is so powerful, way beyond what most people would ever realise. So I wanted to see how far I could take one myself.

Here's what's been rattling round my head. With the speed everything's moving, we jump straight to the end a lot. Ask, get the answer, ship it. Brilliant, don't get me wrong, but where does the thinking go? For me, canvases are where it happens: expanding, diverging, exploring, getting it wrong in five directions before one feels right. Noodling. Hence the name.

What I had to steer

I gave myself permission to go off piste. A bit bold, a bit hacky and properly opinionated. Canvases have a reputation as the place good projects go to get confusing, hard to track and never lead to an outcome, and I wanted to poke at every one of those flaws and see what happens.

My spicy take: most boards shouldn't be long-running, surviving artefacts, and I think the world has come round to that. None of these platforms has ever been able to own true work, so why keep trying to be something they're not meant to be? Get in, do incredible work, then get out with a conclusion: a board summary or outcomes doc written for you, by the tools or by agents.

What I wired together

It's built on tldraw, a company, tool and founder I've admired and followed for ages. They're a big part of why basic whiteboarding is table stakes now, which meant I could skip all of that and spend every hour on the weird stuff. Boards sync live on Cloudflare Workers and Durable Objects, Nood is drawn in three.js with its own little blob shadow, and a board owner can hand an AI assistant a key to just that one board over MCP. Locally my ChatGPT plan can join in as a thought partner. The hosted app runs with no AI at all, and everything still works.

Where I pushed it

My absolute favourite right now is Nood. Wake it up and run it round the board with the arrow keys. It drops notes, picks them up and carries them about, votes and stamps things Decided. Press C and a crowd of tiny visitors wanders over and huddles round whatever's winning. Everyone else on the board gets their own Nood with a name tag, so you can walk up to someone and follow them or pass the pen. Is any of it necessary? Absolutely not. It's still my favourite thing in there.

Then the board itself keeps nudging you towards an answer. The question you're working on stays pinned at the top. Notes get stamped Decided, Parked or Killed. Funnel a pile of ideas down to three. Hit Wrap up and the decisions, next actions and assumptions to test get written into a Doc for you. Say what came out of it, set the board aside and go and do the thing.

Live site

See it live.

noodla.app homepage This site refuses to be framed, on purpose. Open it in a new tab.
Live app Free to play with. Start a board without signing up, or sign in to keep boards private. It won't sit inside another site, so it opens in a new tab.

The build log

How far can a canvas go?

The question I keep coming back to is what thinking even looks like now. Everything’s moving so fast that we skip straight to the end. You can get to an answer in seconds, and most days that’s brilliant. But the good stuff, for me anyway, comes from the mess before the answer: going wide, getting it wrong, smashing two daft ideas together and spotting a third. I wanted somewhere to play with that, properly, and see if my own thinking about thinking moved along with it.

Built on tldraw

v0.1 landed on 26 October and got the basics in: shared boards, notes, Markdown docs and AI answers streaming onto the canvas. It had a placeholder name and no real point of view yet. Building on tldraw meant the whiteboard part was sorted from day one, so every hour after that could go on the opinionated stuff.

Once the noodling idea clicked, it became Noodla, with twisted cable noodles that link ideas together and follow them round the board.

“Add them all and deploy”

On 7 November I brainstormed a load of creative thinking moves, and that was the whole instruction to the agents. Three waves went live the same day.

Tab branches a thought. Ghost branches offer Why?, Example, Opposite and What next?, and the opposite gets linked with a red tension noodle. P deals a provocation card when you’re stuck. There’s a five-minute burst with a countdown ring, lenses that tint the whole board (sceptic, customer, beginner, 10×), magnets that hold clusters together, and a silent round that hides everyone’s notes until somebody hits reveal, so nobody anchors on the loudest person in the room. Then a time-lapse replays the board in the order it was made.

Then Nood

My personal favourite. The brief was “go straight to 3D with little shadows, build and deploy”, and that’s exactly what it is: a little purple mascot sat on a bench in the corner. Wake it up and run it round. It drops notes, picks them up and carries them about, votes and stamps things Decided (it cheers when you do). Leave it alone and it wanders off, sits on notes nobody’s touched for a few minutes, then walks back to its bench and falls asleep. During a silent round it tiptoes.

Everyone on the board gets their own, with a name tag. Walk up to someone and you can follow them, wave or pass the pen.

Boards that land

Here’s where I get opinionated. The question you’re answering stays pinned at the top. Notes get stamped Decided, Parked or Killed. Funnel a pile of ideas down to three and the rest get parked. Wrap up turns all those marks into a Doc of decisions, next actions and assumptions to test. In Direction you say what came out of it, pick the Doc as the result and set the board aside. Nothing’s deleted, you can always bring it back.

I don’t think most boards should live forever. Get in, noodle, get out with a conclusion. Agents can do the getting out too: a board owner can give an AI assistant a key to that one board, read only or read and edit.

Where it’s heading

The ideas list is getting silly in the best way. The plan is multiverse boards you fork at a decision and merge back together, idea breeding (crossbreed two notes, mutate one, let the votes pick the survivors) and a graveyard that resurrects a killed idea when the reason you killed it stops being true. None of that’s built yet.

Will keep you posted.

The screenshots are from a sample board on a local build.