Cr34te.it, a project by Institut für digitale Herausforderungen, has launched a novel platform that allows users to collaboratively edit a website one turn at a time using AI agents. The concept is deceptively simple: users claim a turn, copy a specific prompt into an AI chat or local coding agent, instruct the agent on what to change, and submit the resulting patch. The platform acts as a gatekeeper, ensuring that edits are applied sequentially and safely, turning website maintenance into a gamified, queue-based collaborative exercise.

The Turn-Based Editing Workflow

The core mechanic revolves around a strict queue system. When a user claims a turn, they receive an active slot or a place in the queue. If the queue is empty, editing begins immediately. Users are strongly encouraged to use a local coding agent rather than a web-based chat, as local agents can iteratively discuss changes with the user before submission. Web chats, by contrast, must produce the entire result in one go, offering no preview of whether the patch will apply correctly. Once the AI generates the code, the user pastes the result back into Cr34te.it, where the system performs validation checks before publishing.

Technical Constraints and Safety Mechanisms

To prevent chaos, Cr34te.it enforces rigorous technical limits. Each patch is capped at 30 KB, with a maximum of six changed files and 100 KB per file. The total editable site size is limited to 500 KB, with only 15 files available for modification. Agents can only edit main page files and add CSS or JS files within the assets/ directory. Furthermore, the platform explicitly does not support Gemini, as its web chat cannot open the necessary web addresses to follow the instructions. All submissions undergo safety checks for unsafe text, malicious code, and exploit attempts, with failed patches being rejected outright.

Community Governance and Version History

The platform introduces a layer of community governance through voting and reporting. Signed-in visitors can upvote or downvote the current version, with well-liked versions highlighted in the history strip and disliked ones fading. If enough users report a version, it is automatically removed from history and reviewed by the operator. Editing time is separated from live time; every accepted change receives a guaranteed live window, during which new users can join the queue. For those in a hurry, a 'Skip queue' option allows users to pay via Stripe to jump ahead, though tickets are only processed after payment confirmation.

Key Takeaways

  • Cr34te.it uses a queue-based system to serialize AI-driven website edits, preventing merge conflicts.
  • Local coding agents are recommended over web chats for better iterative control and preview capabilities.
  • Strict file size and count limits (30 KB per patch, 15 editable files) ensure platform stability.
  • Community voting and reporting mechanisms allow for decentralized quality control of site versions.

The Bottom Line

Cr34te.it is a fascinating experiment in collaborative, AI-assisted development that prioritizes order over speed. While the queue system might feel restrictive to power users, it offers a compelling model for safe, incremental website evolution driven by community input. This isn't just a toy; it's a prototype for how we might manage shared digital spaces in the age of autonomous agents.