> ## Content Index > Fetch the complete content index at: https://www.notatechguy.com/llms.txt > Use this file to discover other available public pages before exploring further. # doop open-source canvas lets AI agents design live with humans - URL: https://www.notatechguy.com/doop-open-source-canvas-lets-ai-agents-design-live-with-humans/ - Published: 2026-11-04T08:40:24.000Z - Updated: 2026-11-04T08:40:24.000Z - Description: Open-source design tool doop lets AI agents and humans edit the same canvas simultaneously, with built-in MCP and six model families. - Author: Marcello Babbili - Tags: Technology & AI, AI Agents, OpenAI Developer kgoedecke published doop, an open-source design tool that landed on GitHub's daily trending list today with 1,589 total stars, 125 of them added in the last 24 hours [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com) — figures that are point-in-time and shift quickly. The maintainers position it as a free, self-hosted alternative to Paper.design, built around a premise most design tools still avoid: AI agents and humans editing the same canvas at the same time, live [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). The "alternative to Paper.design" framing is the maintainers' own positioning, not an independent market classification [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). ### Agents and humans share one live canvas Each design occupies a shareable canvas divided into frames, which display live HTML rendered within sandboxed iframes [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). A built-in MCP server allows AI agents to write edits directly into those frames and broadcast their progress as it happens, so a human collaborator sees the agent's changes appear in real time [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). The iframe sandboxing and MCP server access controls are described only in the project README and have not been independently audited [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). The bundled Doop Agent works with six model families: GPT, Claude, Gemini, DeepSeek, Kimi, and Qwen [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). Multiplayer runs through a single WebSocket room that handles live cursors, presence, follow mode, per-frame editing indicators, undo and redo, threaded comments, and an activity feed [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). The project defaults to private access, with collaboration gated by email invites or shareable links restricted to view, comment, or edit permissions [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). ### Self-hosting with zero external services The maintainers state that one `docker compose up` or `bun run dev` command launches the tool with no configuration, relying on an embedded Postgres database and no external services [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com) — a claim that has not been tested outside the project. The codebase is written in TypeScript [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). The first public release, v0.1.0, shipped on 22 August [P²](https://github.com/kgoedecke/doop/releases/tag/v0.1.0?ref=notatechguy.com), and the project has since pushed 383 commits to main [P²](https://github.com/kgoedecke/doop/releases/tag/v0.1.0?ref=notatechguy.com). A desktop build, v0.11.0, was released on 1 November with 14 platform assets including RPM and signed packages [P⁴](https://github.com/kgoedecke/doop/releases?ref=notatechguy.com). For a frontend lead at a small product team, doop would replace a manual handoff step: instead of exporting mockups to a separate tool, the designer and an AI agent could work on the same HTML frames simultaneously, with the agent's edits visible live on the canvas [S¹](https://github.com/kgoedecke/doop?ref=notatechguy.com). The practical question is whether the iframe sandboxing actually isolates agent-generated HTML from the parent application — the single claim that, if it fails, makes the tool unsafe to run with untrusted model output. The desktop v0.11.0 build shipped on 1 November [P⁴](https://github.com/kgoedecke/doop/releases?ref=notatechguy.com) is the most recent artefact, and the 383 commits since the August v0.1.0 release [P²](https://github.com/kgoedecke/doop/releases/tag/v0.1.0?ref=notatechguy.com) are the fastest way to gauge whether development is sustained. --- *Sources: [S1 — kgoedecke/doop: The open-source alternative to Paper.design. A multipl](https://github.com/kgoedecke/doop?ref=notatechguy.com) · [P2 — Release doop v0.1.0 — initial open-source release · kgoedecke/doop · G](https://github.com/kgoedecke/doop/releases/tag/v0.1.0?ref=notatechguy.com) · [P3 — Trending repositories on GitHub today · GitHub](https://github.com/trending?ref=notatechguy.com) · [P4 — Releases · kgoedecke/doop · GitHub](https://github.com/kgoedecke/doop/releases?ref=notatechguy.com) · [P5 — GitHub - huggingface/paperswithcode: This repository hosts the code of](https://github.com/huggingface/paperswithcode/tree/main?ref=notatechguy.com)* --- *Written from 5 sourced items, 4 of them primary.* ![doop GitHub stars: today vs total](https://storage.ghost.io/c/6e/89/6e896869-22ef-4281-a213-b4c462c17cff/content/images/2026/11/chart_407ff16782f730007ed9.png) ## More from Not A Tech Guy - [OpenAI claims ChatGPT speeds up CAD and PCB design](https://www.notatechguy.com/openai-claims-chatgpt-speeds-up-cad-and-pcb-design/) - [FedReady AI agents find all federated learning sites at 65% precision](https://www.notatechguy.com/fedready-ai-agents-find-all-federated-learning-sites-at-65-precision/) - [Prompt injection defense cuts agent attack rate from 37.8% to 4.4%](https://www.notatechguy.com/prompt-injection-defense-cuts-agent-attack-rate-from-37-8-to-4-4/)