---
title: 'DeepSeek + REA Is INSANE'
source: 'https://youtube.com/watch?v=G0WqLjlSXNo'
video_id: 'G0WqLjlSXNo'
date: 2026-11-09
duration_sec: 417
channel: 'Julian Goldie SEO'
---

# DeepSeek + REA Is INSANE

> Source: [DeepSeek + REA Is INSANE](https://youtube.com/watch?v=G0WqLjlSXNo)

## Summary

The video demonstrates how to combine DeepSeek's agent harness with the open-source REA (Reverse Engineer Anything) tool to reverse-engineer desktop apps and websites, then rebuild them as custom business tools. The creator walks through three practical builds: an agent control deck, a lead flow reactor, and a personal website rebuild, providing step-by-step setup and prompts.

### Key Points

- **Introduction to REA and DeepSeek harness** [00:02] — The video introduces REA, an open-source tool with 438,000+ stars, and DeepSeek harness, an agent app with 275,000+ stars, both MIT licensed. REA provides reverse engineering capabilities for native, JavaScript, Electron, .NET, Android apps, and websites, with evidence-based answers.
- **Setup commands** [00:32] — Setup involves running 'npx REA-agents setup -- client dsh' to register REA inside DeepSeek harness, requiring Node 24. Then 'npx rea agents doctor' to verify health. If quota issues arise, add OpenRouter with a key and select DeepSeek v4.1 Flash.
- **Build 1: Agent control deck** [02:45] — Using REA to reverse-engineer Ion UI, an open-source desktop app, the creator built a custom agent control deck for a dental practice with six live agents, light trails, and a command box. REA mapped 38,000 pieces and 548 source files, and honestly reported what it could not read.
- **Build 2: Lead flow reactor** [04:24] — REA reverse-engineered Agent, an open-source app, revealing that task splitting happens in a Python backend. The creator rebuilt this pattern for leads, splitting new leads into tasks (qualify, reply, book, follow-up) across four worker lanes, with a celebration when appointments are booked.
- **Build 3: Website rebuild** [05:21] — REA analyzed juliangoldie.com, launching a real browser to capture structure, fonts, scripts, and buttons. DeepSeek rebuilt it as a premium version with serif headlines, soft light, and clean layout. A revision was requested and fixed in one pass.
- **Lessons learned** [06:05] — Key takeaways: always ask for an evidence file, focus on the app's own code, only reverse-engineer software you own or open-source, and specify design details (colors, fonts, motion) for better results.

## Transcript

rebuild them. I plugged a tool called REA, short for [music] reverse engineer anything, into DeepSeek harness. Then I pointed it at three apps on my [music] Mac. It opened each one up, showed me the evidence of how it works, and
rebuilt the idea as a beautiful business tool, an agent control deck, a lead flow reactor, and my own website rebuilt [music] from scratch. I'll show you every build, every prompt, and the exact setup step by step.
So what is REA? It's an open source tool on GitHub from Morluto. It has over 438,000 stars and it's MIT licensed. It gives your AI agent one set of tools for reverse engineering, native programs,
JavaScript and Electron desktop apps, .NET apps, Android apps, and websites. Everything runs on your own machine. And the part I love, every answer comes with the evidence behind it plus what it could not prove.
Now the brain. DeepSeek harness is DeepSeek's own agent app. It's open source too, MIT licensed with over 275,000 stars. Everything inside it is a plugin, so REA just slots in. I'm running
so REA just slots in. I'm running DeepSeek V4.1 Flash. It's fast and it can look at screenshots which matters when it checks its own work. Here's the setup. Open your terminal and run npx REA-agents setup -- client dsh. That
registers REA inside DeepSeek harness and installs the reverse engineering skill. One thing caught me out. REA needs node 24. If setup stops and asks for node, install node 24 next to your current one and run it again. Then run
npx rea agents doctor. You want the DeepSeek harness line marked healthy. Restart harness, open a new session, pick your folder, and pick your model. If DeepSeek tells you your account quota is used up, go to settings, then models,
add OpenRouter with your key and pick DeepSeek v4.1 Flash from that list. That's exactly what I did for these builds. That's it. Three commands and DeepSeek can now look inside software. Quick one. If you want my best setups
for DeepSeek and REA, like the three prompts from this video, the evidence checklist, and the design brief that made these builds look this good, they're all inside the AI Profit Boardroom. You also get the faceless
video skill I used to make this video, plus a community of people building with AI every day, and a new tutorial almost every day. The link is in the comments and the description, or go to aiprofitboardroom.com.
aiprofitboardroom.com. Build one. The agent control deck. Ion UI is an open-source desktop app that runs several AI agents side by side. I asked DeepSeek to use REA to find out how it does that, then build my own
version for a dental practice. Here's my prompt. Use REA to reverse engineer how Ion UI runs several AI agents side by side. Write the evidence to evidence.md.
Then build my own version, an elegant agent control deck with six live agents, light trails, and a command box. Watch what happens. It loads the reverse engineering skill, unpacks the app, and maps it. 38,000 pieces inside one app
maps it. 38,000 pieces inside one app file. 548 source files checked. Then it hit a limit. Two bundles were too big to pass, so it raised its own memory limit, narrowed the search to the app's own code, and kept going. And here's the
honest part. Its report says exactly which conclusions came from the code and which parts it could not read. It found the agent manager at the heart of the app, one adapter for each kind of agent, and how each conversation tracks its
and how each conversation tracks its status. Then it built this. Six live agents for Bright Smile Dental. Lead finder, inbox, SEO writer, booking, support and reports. Each card streams what it's doing. Light trails flow from
the DeepSeek lead to every agent. And when I type a new task, it flies to the right agent, and that agent picks it up. Build two, the lead flow reactor. Agent is another open source app. You give it a job, and it splits that job across a
team of worker agents. I asked DeepSeek how the splitting works. My prompt, use REA to find out how Agent breaks a request into tasks and hands them to worker agents. Write the evidence. Then build a lead flow reactor for my dental
practice. This is where REA earned its place. The desktop app turned out to be just the front. The real task splitting lives in a Python backend shipped inside the app. DeepSeek would never have guessed that. REA found it. So it
rebuilt that pattern for leads. New leads fly into the reactor. Each one leads fly into the reactor. Each one gets split into tasks, qualify, reply, book, and follow up. The tasks shoot out to four worker lanes. And when an
appointment gets booked, the whole screen celebrates. I can add a lead by hand, and it goes straight through the same flow. Build three, my own website same flow. Build three, my own website rebuilt. My prompt, use REA's website
analysis on juliangoldie.com, which I own. Map the sections, fonts, scripts, and buttons. Then rebuild it as an elegant premium version. This time, REA did something different. It launched a real browser, loaded juliangoldie.com,
and captured the page structure, the fonts, the scripts, and where every fonts, the scripts, and where every button leads. 15 sections mapped. Then DeepSeek rebuilt it with my real sections and words. Big serif headlines
that fade in as you scroll. Soft, warm light. Every service, every result, and real reviews laid out cleanly. The first version copied a banner I didn't want. So I asked for a revision right in in same chat, and it fixed it in one pass.
Here's what I learned from these three builds. One, always ask for an evidence file. It keeps deep seek honest, and you can check every claim. Two, on big apps, tell it to focus on the app's own code, not the bundled libraries. Three, only
look inside software you own or open source software you're allowed to study. Then build your own version with your own name and design. Four, ask for the look in detail, colors, fonts, motion. That's the difference between a plain
dashboard and something people actually want to use. If you want the exact prompts for all three builds, the setup checklist, and the design brief, they're inside the AI Profit Boardroom along with [clears throat] the faceless video
skill that made this video. The link is in the comments and the description, or go to aiprofitboardroom.com. Thanks for watching, and I'll see you in Thanks for watching, and I'll see you in the next one.
