tldraw computer

Developed by the tldraw team in partnership with Google's Gemini AI, the project demonstrates a fundamentally different paradigm for AI interaction: one where data flows visually between connected components on an unlimited canvas, and where complex multi-step AI workflows can be constructed and understood as spatial diagrams rather than nested prompts.

The architecture is built around components discrete elements on the canvas representing inputs, transformations, or outputs linked by arrows that visualize how information moves through a workflow.

Text boxes, images, audio clips, code blocks, and AI model calls can all be placed as components and wired together in custom configurations.

The natural language instruction layer means users describe what they want each transformation to do rather than writing code, making it possible for non-developers to construct sophisticated data processing and content generation pipelines by drawing connections and typing descriptions.

The Gemini integration powers the natural language understanding that makes tldraw computer functional as a visual programming environment.

When a user types an instruction like 'summarize this document and identify the three main arguments', Gemini processes the connected input component, performs the requested operation, and writes the result to an output component all visualized as a node in the canvas diagram.

More complex workflows chain multiple AI operations, such as fetching web content, summarizing it, translating the summary, and generating an image based on the translated output, with each step visually represented as a connected node.