tldraw Launches an Offline Desktop Version: Infinite Canvas and AI Agents Move to the Local Workstation

What happened

tldraw released the offline desktop app tldraw offline, bringing its infinite canvas from the browser to macOS, Windows, and Linux. The app is built around a local-first approach: users do not need to sign up for an account or rely on remote servers, and can draw, diagram, and edit whiteboard documents even without an internet connection.

The app is built on the tldraw SDK. According to the official introduction, the same SDK is also used by tldraw.com and by hundreds of companies developing canvas-based products.

Why it matters

Documents are saved as portable local files

tldraw offline uses self-contained .tldraw files to save documents. These files can include canvas content, pages, media assets such as images and videos, and any embedded document scripts.

The official documentation defines .tldraw as the app’s native and preferred file format. Files can be used across macOS, Windows, and Linux, and users can also back them up themselves, store them privately, or share them directly with others.

This makes tldraw offline more than just an offline whiteboard client; it also offers a way to manage documents without depending on accounts or cloud servers.

AI agents can directly participate in canvas creation

Another key capability of tldraw offline is that it allows AI coding agents to view and modify the canvas. The agents listed by the official materials include Claude Code, Codex, Pi, and OpenCode.

These agents can read the current document, modify content in the editor, and write reusable scripts for the canvas. The GitHub project description also notes that the desktop app runs a local HTTP service and provides a Canvas API for programmatic access to tldraw documents, supporting AI coding assistants and other tool integrations.

This means AI is no longer limited to generating text or code in a chat window; it can also enter a concrete visual workspace. Users can first sketch out diagrams, flowcharts, or interface structures on the canvas, and then let an agent read the canvas state and continue making edits.

Evidence and confirmed information

• tldraw offline supports macOS, Windows, and Linux.
• The app requires no account or remote server and can be used offline.
• Documents are saved as self-contained .tldraw files, which can be used across the above platforms.
• Agents such as Claude Code, Codex, Pi, and OpenCode can read and modify the canvas and write document scripts.
• The GitHub project describes a local HTTP service and Canvas API for programmatic access to tldraw documents.
• The official reminder says that agents granted access to the app can read and edit user documents; scripts in .tldraw files may run when the file is opened. Therefore, users should grant access only to trusted agents and open only files from trusted sources.

What to watch

First is the permission boundary for AI agents. Because agents can directly read and edit local documents, and embedded scripts may also run when files are opened, users need to carefully manage agent permissions, file sources, and local backups.

Second are the differences between the desktop version and tldraw.com in terms of features, collaboration methods, and document compatibility. The current materials confirm offline use, local files, and AI agent capabilities, but they do not provide full performance tests or an official release timeline.

In addition, the tldraw community has already seen extension practices around AI and local canvases. For example, Agent Draw lets users select a region of the canvas and describe their needs by voice, after which an AI agent draws charts or illustrations; other projects try to drive the canvas through the command line or JSON-RPC. These projects are not official features of tldraw offline, but they show that the tldraw canvas model is being used to build local AI workflows.

Overall, the core change in tldraw offline is not merely that “you can draw without the internet,” but that it combines canvas, portable files, and AI agents into a local working environment. Whether it can further become a commonly used tool for flow design, product prototyping, architecture diagrams, and visual brainstorming will depend on the real-world performance of its agent integration, platform consistency, and script safety mechanisms.

Original source

tldraw offline official website