tldraw

Notes now centre exactly where you click inside a rotated frame, instead of landing off to the side.

In case you missed it, we made tldraw flash, a free animation tool built on the tldraw SDK. You can sketch something, hit record, and drag it around the canvas to animate it. You can trim clips on the timeline, add sound or a voiceover, export to MP4, or publish to the gallery so others can remix it. You can also: → Bring in your own images and cutouts and move them around like stickers. → Draw a character in one scribble, and it gets a skeleton. Drag it, and it walks with you. → Remix anything in the gallery. → An AI agent can drive it too. Someone on X made a five-act, 2:52 Odysseus short from 1,863 tldraw shapes. Haven't tried it yet?

Teamwork makes the dream work

We’re pleased to welcome Canary (株式会社カナリー) on board as our latest customer! Canary uses tldraw’s infinite canvas SDK to create easy-to-read floor plans for rental listings. tldraw’s custom shapes allow users to model rooms as polygons, snap walls and fixtures to the 910mm Japanese architectural grid, and support multi-floor plans. With tldraw, Canary can focus on the rental market instead of reinventing the canvas.

2D space is the next frontier for agents. Unlike chat, 2D spaces like canvases give you an interface to jam with agents like you would with another person: point things out, give annotated feedback, and sketch ideas on the fly. But for agents to operate in a 2D space, they need to be told how to move around it. For that, we built something we call the Spatial Harness. It evolved over multiple demos and products, from Teach, to Fairies, to tldraw offline. Max Drake, a tldraw Product Engineer, dives deep into how we built it, and what it means for agents on the canvas. Read the full story in the comments.

WorkOS founder Michael Grinich brought together builders from tldraw, Attio, Daytona and incident.io for a night of live demos showing what agentic AI already looks like in real products. A strong night for London’s founder and developer community. #AI #AgenticAI #LondonTech #Founders

Double-clicking a frame's edge to fit its content used to shift everything inside it. In the new version of tldraw.com, the frame moves to compensate, so shapes stay put.

Holding Z to zoom sometimes flipped the toolbar to the wrong tool, or left it stuck on zoom. Now, it stays on the tool you picked.

We spend a lot of time building on the canvas, as well as writing about it. Lately, that's included running AI models inside a tldraw offline file, teaching doodles to walk, and figuring out how to tell an agent "this bit here, not there". Here are a few pieces from the archives, in case you missed them: 🔗 Text is not enough Understanding an AI's answer well enough to question it can take some work, especially when the answer is about something visual. Steve Ruiz shows what happens when you give the conversation a shared canvas. An agent can draw two approaches side by side; you can edit the drawing and use it to steer the next step. Handy for working through a design, a process, or a change to a codebase. Read the post → https://lnkd.in/ec5bahAx 🔗 Agents can't point "The blue rectangle. Top left. No, the other one." Tim Gilboy tried having an agent leave feedback as comments pinned to a screenshot on the canvas. Each suggestion had a place, a thread, and somewhere to reply. It's a small but useful detail in designing tools for working with AI: keeping the conversation attached to what you're talking about. Read the post → https://lnkd.in/eE5c9f3a 🔗 Building with tldraw offline Good starting points for making software around the way you actually work. We collected tools and experiments built with tldraw offline, including examples you can download and change with a coding agent. Read the post → https://lnkd.in/eh9dPVf9 🔗 Doodling across disciplines Daniel Lefcourt explains how a technique from 3D games helped us make ordinary scribbles bend and walk in tldraw flash. Useful reading if you care about making creative tools that fit how people think. Read the post → https://lnkd.in/eerj4hed What should we write about next?

It’s been a busy week in the canvas and design dev space on X. Here are a few things that got us talking in the tldraw office: → The 45-second AI explainer trend. Matija Šošić Sosic, CEO of Wasp, kicked it off with a with a "45-second TL;DR" video explaining Jev. The reaction was split: some loved the compression; others felt that cramming a complex idea into 45 seconds of AI video can make it harder to follow, not easier. Ani Krishnan, a tdraw Product Engineer, made the format his own. Instead of asking an AI to generate a video, Ani had an agent draw the explanation live in tldraw offline, stroke by stroke. → AI models directing their own art. @devteamdrew from Lemonade set Claude Opus 5 loose on one open-ended goal — build an infographic of the water cycle. Four hours later, it had produced a finished four-minute MP4, every visual decision Claude's own. Days later, @itssssDean, a video content creator, had Opus 5 write a butterfly's life cycle entirely in JavaScript: 17 shots, ~768 frames, no diffusion model, just code drawing frame by frame. He liked the process enough to hand the same prompt to Kimi K3 and compare notes. → Sora's API goes dark September 24. The Sora app closed back in April, and the API winds down next week. Teams wired to it are migrating this month, mostly to Runway, Luma, and Veo, and mostly through hubs like fal, Krea, and Higgsfield that bundle several video models behind one workflow. If you build creative tools, it’s worth knowing that these are almost all node-canvas products. A lot of thought is happening right now about how to rebuild video-gen pipelines, and what that could feel like on the canvas. Watch this space for our take. Links to threads in the comments. What got your office chatting this week? We’d love to hear it.