Rebuild an open-source Grok Bot with parallel cloud agents

David runs several Claude Code agents in Cloud Room to reverse-engineer Grok Bot, build its UI and back end, and self-host the result on a VPS in about 1.5 hours.

David Ondrej · YouTube · published Nov 15, 2026 · Software & DevOps

Rebuild an open-source Grok Bot with parallel cloud agents — video by David Ondrej▶ Watch on YouTube
Prompt
We're building an open-source version of [the app I want to copy]. Set up this repo with a basic README and structure, then ask me a few concise questions. Answer in short, plain English.

Visual breakdown

David rebuilds the open-source version of Grok Bot in about an hour and a half by running several coding agents at once in Cloud Room: one researches, one builds the UI, one the back end, one deploys it to his own VPS. Claude Code CLI is the harness under the hood.

The problem: Grok Bot is closed, so you can't self-host it, swap its model or see how it works under the hood.

How he rebuilt it with a team of agents

  1. Screenshots + Grok's own answersHe asks Grok Bot to describe its VM and architecture, then pastes that and UI screenshots to agents
  2. Research and docs agentsDocument the internals as modular markdown files and save the architecture to the README
  3. UI agent (cloud)Builds a look-alike UI from the screenshot and runs it on localhost
  4. Back-end agentBuilds the back end on top of Claude Code CLI: supervisor, scheduler, message bus, guardrails
  5. Hostinger VPS + SSH keyBuys a VPS, adds an SSH key, lets an agent lock it down and set up auto deploy from main
  6. Review and iterateModel selector, bot creation, custom icons, inter-bot messaging, GPT review by a second harness
    Dropped: Weak designs get four new options
  7. SuperbotOpen-source, self-hosted Grok Bot alternative powered by his own Claude subscription
InputsSetup / manualAI agentsResult

Habits he repeats while building

  1. Start with several agents at once, not one
  2. Ask for short, plain-English answers (agents.md rule)
  3. Get it deployed early and test in production
  4. Use "don't overthink it" so agents stay practical

By the numbers

~1.5 hfrom empty repo to deployed
1%of his Claude Code limit used early on
67%of the repo is Rust
41agent loops he says he runs

Try it yourself

He works in Cloud Room (his open-source coding app), but the steps are the same with any agent setup.

  1. Create a new repo and ask one agent for the basic README, structure and a few questions
  2. Start more agents: one to research and document how the app works, one for the UI from a screenshot
  3. Push to GitHub early and keep a short agents.md telling agents to answer concisely
  4. Ask an agent to build the back end on top of a CLI harness such as Claude Code
  5. Get a VPS, add an SSH key, and have an agent secure it and set up auto deploy from main
  6. Iterate in small asks: model selector, new designs, icons, then review with a second harness

The finished repo is called Superbot and is linked below the video.

Tips and caveats from the video

  • Agents answer at length unless you tell them to be short, and will overbuild without "don't overthink it".
  • Right-size the reasoning effort: high for small tasks, extra high or max for big ones.
  • Keep notes on trademark: he renamed the project to Superbot at the end.
  • He says you still decide what to build; agents are weak at naming and creative choices.
  • Anthropic's policy on third-party harnesses was unclear to him, so he used Claude Code CLI itself as the harness.

Distilled from the video’s captions

How to try it

  1. Open Claude Code.
  2. Copy the prompt and fill in the [placeholders].
  3. Watch David Ondrej's video for the full walkthrough and setup.

Related use cases

Share