YT YouTube2Text

YouTube2Text → Transcripts → Forget Claude Code: DeepSeek + REA Clones Anything 100% FREE (No Paid API)

Forget Claude Code: DeepSeek + REA Clones Anything 100% FREE (No Paid API) — Transcript

by Maitis · 3,817 words · 583 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00When Apple launched the iPhone in 2007,
0:02Google was secretly building a phone
0:04with a tiny screen and a keyboard. They
0:07took one look at the iPhone, studied how
0:09it worked, and rebuilt their whole phone
0:12around a touchscreen.
0:13That became Android, and today it runs
0:16on most of the phones on the planet,
0:18maybe even the one in your hand. Google
0:20never needed Apple's code. They just
0:22understood how it worked and built their
0:24own.
0:25But Google had thousands of engineers to
0:27do that. For people like you and me, the
0:30apps we use are still locked boxes. We
0:33can use them, but we can't see how
0:35they're built, and every month we pay
0:37again.
0:38So, what if you didn't need a team of
0:40engineers? What if you could look inside
0:42any app and build your own version
0:44without paying anything? Two weeks ago,
0:47I showed you how to clone any app with
0:49Claude code, REA, and Nine Router. It
0:52worked, but Claude burned through the
0:54credits way too fast. So, today we're
0:57going to explore a free workflow with
0:59DeepSeek, REA, and OpenRouter, which
1:02lets you use powerful AI models at zero
1:04cost per token. And to prove it works,
1:07we'll build free clones of two apps
1:09millions of people use every day. Drop
1:12your guess in the comments. And I've
1:14saved the biggest surprise for the end.
1:17People already pay good money for apps
1:19like these, and I'll show you how to
1:20turn this into a service. We're doing
1:23this in three levels. In level one,
1:26we'll set up DeepSeek with REA. In level
1:28two, we'll connect OpenRouter. And in
1:31level three, we'll build those two
1:32clones, and that's where things get
1:35insane. But before we jump in, give me
1:3830 seconds. If you're building with AI,
1:41the idea is never the hard part. The
1:43setup is. One missed command and nothing
1:46works.
1:47>> [music]
1:47>> That's exactly why I built midas.ai.
1:50Every video I make turns into a
1:52step-by-step guide there, with the
1:54commands, prompts, and links in order,
1:56ready to copy and paste. The REA setup I
1:59showed you 2 weeks ago is already up.
2:01And today's Deep Seek and Open Router
2:03workflow goes right next to it. When
2:05you're ready to explore more, there are
2:07760 plus AI tools waiting. And you can
2:11filter for the free ones. And the best
2:13part? It's completely free. So, if
2:16you're building along with me today, the
2:17link's in the description.
2:21And if you're new here, hit subscribe
2:23because we break down stuff like this
2:24every day. All right, let's open the
2:26first box.
2:30Before we build anything, we need
2:32someone to do the actual building. And
2:34we're not hiring a developer, we're
2:36using one. It's called Deep Seek
2:38Harness. Think of it as a developer that
2:41lives on your computer. You tell it in
2:43plain English what you want and it
2:45writes the code, fixes the errors, and
2:47puts everything together for you. So,
2:50let's get it set up. Search for Deep
2:52Seek on Google, open the official
2:54website, and click on Harness Desktop.
2:57Download the app for your computer. I'm
2:59on a Mac, so I'll just drag it into
3:01applications and open it.
3:02>> [music]
3:03>> When it opens, you'll see two options:
3:05sign in and add a PI key. We're going
3:08with the API key.
3:10Now, an API key sounds technical, but
3:13it's really just a password that lets
3:14the app talk to Deep Seek's brain. To
3:17get one, search for Deep Seek API, open
3:20the Deep Seek platform, and sign in with
3:22Google.
3:23Go to API keys, click create new API
3:26key, give it a name, [music] and copy
3:28it. Come back to Harness, click add API
3:31key, paste it in, and continue. [music]
3:34And just like that, we're inside. Let's
3:37take a quick look around.
3:38In settings, you'll find the appearance,
3:41the models, built-in plugins, and agent
3:43presets. We'll use the standard preset
3:46for this project. Now, here's the part
3:48that matters. Harness lets you pick
3:50which AI model does the thinking. We've
3:53got Deep Seek V4.1 Flash and Deep Seek
3:56V4 Pro. And you can even choose how hard
3:59it thinks. For our project, we're
4:01setting the thinking effort to high. But
4:03remember, our whole goal is to do this
4:06without paying for expensive AI. So, in
4:08the next level, we're plugging in Open
4:10Router, which gives us models we can use
4:12for free.
4:14Hold that thought, because right now
4:16there's a bigger problem. Our developer
4:18is smart, but if we just say, "Build me
4:20a copy of this app," it has no idea how
4:23the original actually works. It would be
4:25guessing. And guessing is how you end up
4:28with something that looks right, but
4:30falls apart the moment you click a
4:31button.
4:32That's where REA comes in. REA stands
4:35for reverse engineer anything. Think of
4:38it as a detective's toolkit for
4:40software. On its own, it's just a box of
4:43tools sitting on a shelf. But plug it
4:45into your AI, and the AI becomes the
4:47detective. It can open up an app, study
4:50how it's built, and trace exactly how
4:52each feature works. Desktop apps,
4:55Electron apps, even compiled programs,
4:58REA has a tool for each of them. So,
5:00instead of guessing, our developer
5:02actually investigates first and then
5:05builds. And REA isn't tied to one AI. It
5:08works with Claude Code, Gemini CLI,
5:11Cursor, and a bunch of others. Today,
5:14we're plugging it into Deep Seek
5:15Harness. By the way, if you want the
5:18full deep dive on REA, we've already
5:20made a complete video on it called, "You
5:22Can Clone Any App for Free: REA Plus
5:25Nine Router, A Complete End-to-End
5:27Tutorial." The link is in the
5:29description. Now, let's plug the toolkit
5:32into our developer. Search for REA on
5:34Google and open its GitHub page. Scroll
5:37down to the readme, where you'll see its
5:39features, supported apps, and the
5:41install instructions. Open your terminal
5:43and paste this one line: npx
5:46REA-agents@latest-setup-client-dsh.
5:51Hit enter. That one line tells REA, "Set
5:55yourself up for Deep Seek Harness."
5:57You'll see a short plan of what it's
5:58about to install. In our case, that's
6:01the REA skill, the harness settings, and
6:03an extra analysis tool called Hopper.
6:06Think of Hopper as a magnifying glass
6:08for apps that are harder to read. Type Y
6:11and let it run. Once it's done, we need
6:13to make sure everything actually works
6:15and REA has its own health check for
6:18that. They literally call it the doctor.
6:21NPX REA {dash} agents doctor. The doctor
6:25checks every tool, every connection and
6:27tells you if anything is missing. You
6:29want it all green. And if something
6:31isn't green, here's the trick. Don't try
6:34to fix it yourself. Copy the error,
6:36paste it into Harness and say, "Fix
6:38this." That's how you do technical
6:40setups without being technical. Last
6:43step, restart Deep Seek Harness. Just
6:46like your phone after an update, it
6:47loads its new tools when it starts up
6:49again. So, in plain English, here's what
6:52we've got so far. Deep Seek Harness is
6:54our developer, the one who writes the
6:56code. REA is the detective's toolkit
6:58that lets it look inside other apps. And
7:01the doctor makes sure it's all
7:02connected. And if you've made it this
7:04far with your setup working, quick
7:06favor. Hit subscribe because we break
7:08down tools like this every single day
7:11and you don't want to miss the next one.
7:13But, there's still one piece missing.
7:15Our developer needs a brain to think
7:17with and we don't want to pay for an
7:19expensive one. That's exactly what we're
7:21fixing next and this is where the free
7:24AI models come in. Level two, setting up
7:26Open Router.
7:28So, now we've got our developer, Deep
7:30Seek Harness, and we've given it REA,
7:33the most important toolkit it needs to
7:35study other apps. But, there's still one
7:37thing missing in this setup. We need an
7:39AI model to actually power our developer
7:42and we don't want to keep paying for
7:44every little task. That's where Open
7:46Router comes in. Think of Open Router as
7:49a place where you can access different
7:51AI models through one connection. And
7:53the best part? Some of these models are
7:55free to use. So, let's connect it to our
7:58setup. First, open openrouter.ai,
8:01click sign up and continue with Google.
8:04Select your account, complete the
8:06verification and choose individual.
8:09Now, OpenRouter will give you an API
8:11key. Copy it and keep it safe, and don't
8:14share it with anyone. We'll need it in a
8:16second. You'll also see an option to add
8:18a payment method, but since we're using
8:20free models, we'll skip that for now.
8:22And just like that, we're inside.
8:25Now, let's head back to Deep Seek
8:26Harness. Open settings, go to models,
8:29and add a custom provider. We'll call it
8:32OpenRouter free. For the base URL, paste
8:35this. Then, add the API key we just
8:38copied. And there we go. Our developer
8:40can now connect to OpenRouter. But
8:43before we move ahead, let's quickly
8:44check whether REA is working properly.
8:47Remember, we installed REA in level one,
8:50but we haven't actually tested it with
8:51our AI yet. So, open a new chat, select
8:55OpenRouter/free,
8:56and paste this prompt. Now, let's watch
8:59what happens.
9:00Deep Seek starts checking the
9:01connection, looks through the available
9:03tools, and gives us a report. It even
9:06opens a local system file to check
9:08whether REA can actually inspect
9:09something. And it works perfectly. Now,
9:12before moving forward, we need to pay
9:14attention to a few additional tools,
9:16including Hopper, but the main
9:18connection is working. So, we've got our
9:21AI developer talking to REA, and both
9:23are connected through OpenRouter. Pretty
9:25cool, right? But there's one more thing
9:28I want to show you, and this is actually
9:30one of the most useful parts of this
9:31entire setup. See, free AI models
9:35usually come with limits. Imagine you're
9:37halfway through building an app, and
9:39suddenly the model you're using reaches
9:41its limit. Now, you have to switch to
9:42another model, reconnect everything, and
9:45figure out where you left off. That's
9:47exactly what we want to avoid. So,
9:50instead of relying on one model, we're
9:52going to create a small team of free
9:54models that can back each other up. And
9:56for that, we'll use OpenRouter's
9:58presets. Think of a preset as a saved
10:01setup. We choose the models once, save
10:03everything together, and connect that
10:05setup to Deep Seek Harness. So, head
10:08back to OpenRouter, open presets, and
10:10click new preset. I'm going to name ours
10:13free models. And in the description,
10:16I'll write free AI model fallback for
10:18DeepSeek plus REA. Now, scroll down to
10:21models and click add model. First, we're
10:24adding NVIDIA NeMo-3 Ultra, free. Next,
10:28NVIDIA NeMo-3.5
10:30SuperVL, free. And finally, NVIDIA
10:34NeMo-3.5
10:36Lightning, free. [music] So, now we've
10:38got three free AI models in one place.
10:41And here's why we're doing this. Think
10:43of it like having three people working
10:45on the same project. If one person can't
10:47continue, another can step in. It's a
10:49similar idea here. With model fallback
10:52configured, if one model becomes
10:54unavailable or hits a supported usage
10:56limit, OpenRouter can try another model
10:59from our list. That means we don't have
11:01to manually change our model settings
11:03every single time something goes wrong.
11:05Of course, this still depends on the
11:07available free models and their limits,
11:09but it makes the whole setup a lot more
11:11flexible. Now, scroll down to the tools
11:14section. Here, we'll enable web search,
11:17date time, fusion, advisor, and sub
11:20agent. Once that's done, click save
11:22preset. And OpenRouter gives us a preset
11:25ID, @preset/free-models.
11:30Copy this because we're going to connect
11:31it to DeepSeek Harness. Now, head back
11:33to Harness, open settings, and go to
11:36models. Under our OpenRouter free
11:38provider, click add model. Paste
11:41@preset/free-models
11:44into the model ID field, give it the
11:46display name free models, and click
11:49apply. And that's it. We've connected
11:51our custom model setup to DeepSeek
11:53Harness. So, now, instead of depending
11:56on just one AI model, our developer has
11:58a list of models it can use with
12:00fallback options if one isn't available.
12:03And that's what makes this setup so
12:05useful. Now, let's quickly look at
12:07everything we've done so far. We've set
12:09up DeepSeek Harness as our AI developer,
12:12connected REA to help it understand
12:14other apps, and added open router to
12:16power everything with free AI models. So
12:19now, we've got our developer, the right
12:21tools, and the AI models, all without
12:24needing a paid subscription. And if you
12:26want to discover more free AI tools and
12:28workflows that can replace expensive
12:30ones, subscribe to Midas. We're always
12:33testing them, so you can get more done
12:35without spending money. Now comes the
12:37real test. Can this setup actually build
12:40something useful? In level three, we're
12:42going to recreate the two apps I showed
12:44you earlier and see how close we can
12:46get. Let's put it to work. Now comes the
12:49most exciting part. We've got everything
12:51connected, and it's finally time to see
12:53what this setup can actually build. And
12:56for our first challenge, we're going to
12:58recreate Slack. Now, if you haven't used
13:01Slack before, think of it as WhatsApp,
13:03but for work. It helps teams
13:05communicate, create separate channels
13:07for different projects, share files, and
13:10keep everything organized in one place.
13:12Pretty useful when you're working with a
13:14team and don't want important messages
13:16getting lost between different
13:17conversations. So, here's what we're
13:20going to try. We'll use the free AI
13:22models we just connected to build our
13:24own version of Slack. And not just
13:26something that looks similar. We want to
13:28actually send messages, create channels,
13:31and interact with the app. Let's see how
13:33close we can get. First, open Deep Seek
13:36Harness and make sure our open router
13:38preset is selected. Now, in the chat
13:40box, we're going to give Deep Seek a
13:42detailed prompt explaining exactly what
13:45we want to build. I'm calling our app
13:47Metis Workspace.
13:49We'll ask it to use Slack's publicly
13:50visible interface as a reference, but
13:52with our own name and branding. We want
13:55the familiar purple sidebar, different
13:57channels, direct messages, a messaging
14:00area, and a search bar. And of course,
14:02working features like sending messages,
14:05creating channels, replying in threads,
14:07and sharing files.
14:09We'll also ask it to pay attention to
14:10the smaller details, like spacing,
14:13icons, and how the buttons respond when
14:15we click them. Because there's no point
14:18in building a beautiful app if we can't
14:19actually use it, right? I'll leave the
14:22full prompt in the description, so you
14:23can try it yourself. Now, let's see what
14:26happens.
14:27DeepSeek starts planning the project,
14:29creating a folder, and setting up all
14:31the files it needs. Then, it begins
14:33putting together the sidebar, messaging
14:36area, and thread panel, while connecting
14:38the different buttons and features. And
14:40the interesting part is that we're not
14:42writing any of this code ourselves.
14:44DeepSeek is handling the build based on
14:46our instructions. After a little while,
14:49the build is complete. So, I ran it in
14:51the browser to see what we'd actually
14:53built. And look at this. We've got the
14:55purple sidebar, different channels,
14:58direct messages, employee profiles, and
15:00a complete messaging interface. There
15:03are even message reactions and a
15:04separate panel for replies. It actually
15:07looks pretty close to what we wanted.
15:09But, here's the real test. Does it work?
15:12Let's find out. First, let's click on
15:14the engineering channel. And there we
15:17go. The conversation changes, and we can
15:19see messages from a different team. Now,
15:22let's create our own channel. Click add
15:24channels, type quarterly planning, and
15:27hit create. And just like that, our new
15:30channel appears. Let's send a message
15:32asking the team to share their
15:33priorities for next quarter. And there
15:36it is. Our message appears along with a
15:38sample reply. We can even reply to a
15:40specific message by clicking the reply
15:43icon, which opens a separate thread on
15:45the right. Now, let's try something
15:47else. I'll attach meeting-notes.txt,
15:51add a short message, and send it. And
15:53there we go. The file appears in our
15:55conversation, and we can find it in the
15:57files tab, too.
15:59Moving on, we can also send direct
16:01messages to teammates, open a huddle for
16:03audio conversations, and search for
16:05channels like quarterly planning.
16:07There's even a notes tab where we can
16:09keep track of project priorities and
16:11next steps. So, quite a few features are
16:14working already. And just to make sure
16:16this wasn't a one-time thing, we ran
16:18another quick test. We created another
16:21channel called launch planning, sent a
16:23message, attached a file, opened a
16:25thread, and tested search and huddle
16:27again. And the main interactions worked
16:30this time, too.
16:31Now, of course, this isn't the complete
16:33Slack app. It's a local version with
16:35sample conversations, not a fully
16:38connected messaging platform. But think
16:40about what we've just done. We started
16:42with a prompt, let DeepSeek handle the
16:44building process, and ended up with an
16:46app where we can actually interact with
16:48different features. And we did it using
16:51the free AI models we connected earlier.
16:53Now, Slack was just our first challenge.
16:56We've got another app to recreate, and
16:58this one's going to be just as
17:00interesting. So, let's see how our setup
17:02handles the next one. And this time,
17:04we're taking on something a little more
17:06challenging. We're going to build our
17:08own version of Chat GPT.
17:11Now, we all know Chat GPT. You type a
17:14question, it gives you an answer, and
17:16you can have entire conversations with
17:18it. But there's a lot more happening
17:20behind that simple chat box. So, for our
17:23second challenge,
17:24>> [music]
17:24>> we're going to see if DeepSeek can help
17:26us build something similar with a clean
17:28chat interface, conversation history,
17:31and all the features that make it easy
17:32to use. And the best part? We don't need
17:35to set everything up again. We're using
17:37the same DeepSeek Harness, REA, and Open
17:40Router setup that we used to build
17:42Slack. So, let's get started. Just like
17:45we did earlier, open DeepSeek Harness
17:47and make sure our Open Router preset is
17:49selected. Now, remember how we gave
17:52DeepSeek a detailed prompt to build our
17:54Slack workspace? We're going to do the
17:56same thing here, but this time, we're
17:58asking it to build a Chat GPT-style
18:00application. Clean chat interface,
18:03conversation history, a message box, and
18:06a responsive layout that works on
18:07different screen sizes. And I'll leave
18:09the complete prompt in the description,
18:11so you can try it yourself.
18:13Now, DeepSeek starts checking our
18:15workspace and REA connection, planning
18:17the project, and creating the necessary
18:19files. It builds the sidebar, message
18:22input, and chat interface without us
18:24having to write the code ourselves. Once
18:26it's ready, I ran it in the browser to
18:28see what we'd built. And look at this.
18:31We've got our own ChatGPT style app with
18:34a sidebar, a new chat button,
18:36conversation history, and a message box.
18:39But, let's see what it can actually do.
18:41First, let's ask it to explain
18:43photosynthesis in three simple steps.
18:46And here's where things get even more
18:47interesting. We can generate both images
18:50and videos using just a simple text
18:53prompt right inside our chatbot. For
18:55example, let's ask it to create a
18:57peaceful sunset over a lake. And there
19:00we go. It generates the image based on
19:02our description. We can even create
19:04videos the same way. Just describe what
19:07we want, and the AI can generate it for
19:09us. And the best part? All our
19:11conversations are saved in the sidebar,
19:13so we can come back to them anytime.
19:15Now, let's explore some more features.
19:18Click the plus icon beside the message
19:20box, and you'll see quite a few options.
19:22First is search the web, followed by
19:25deep research and analyze a URL. Then,
19:28we've got create an image, create a
19:30video, add photos and files, my media,
19:34and upload images or videos. Let's
19:36quickly see what some of these actually
19:38do. We'll start with search the web and
19:40paste a Wikipedia link about news
19:42television channels around the world.
19:44And look at that. It organizes the
19:46channels by region, including Asia,
19:49Europe, Africa, and the Americas, along
19:52with source links. Now, using the same
19:54link, let's try deep research. This
19:57time, it gives us a more detailed
19:59report, looking at the reliability of
20:00the information and where there might be
20:02gaps. Next, let's try analyze a URL. And
20:06here, we get a shorter summary of the
20:08webpage, covering its main points and
20:10important details. So, instead of
20:13reading through the entire page, we can
20:15quickly understand what it's about. Now,
20:17moving on to media. We can upload a
20:20reference image, add a prompt, and use
20:22the image creation option. There's also
20:24my media where our uploaded pictures and
20:27videos are organized in one place. And
20:29with create a video, we can enter a
20:31description of the video we want to
20:33make. Now, the image and video creation
20:35options are part of our interface,
20:37although we haven't demonstrated
20:39completed AI-generated media here. But
20:42look at what we've built. We can have AI
20:44conversations, save our chat history,
20:47search the web, get research reports,
20:49summarize web pages, and manage media.
20:52Of course, this isn't a complete
20:54replacement for ChatGPT, and some
20:56features still need further testing. But
20:59we started by building our own Slack
21:01workspace, and now we've created a
21:03ChatGPT-style app using the exact same
21:06setup. Two completely different
21:08applications built without manually
21:10writing the code. And that's what makes
21:13this workflow so interesting. You can
21:15start with an idea, turn it into
21:17something you can actually use, and keep
21:19improving it as you go.
21:22And that's all three levels. We started
21:24with Deep Seek Harness, connected REA,
21:27and used Open Router to power everything
21:29with free AI models. Then we put the
21:32entire setup to the test by building our
21:34own versions of Slack and ChatGPT
21:37without writing the code ourselves or
21:39needing a paid AI subscription. And the
21:41best part? You're not limited to these
21:43two apps. You can take an idea, describe
21:45what you want in plain English, and let
21:47these models help you turn it into
21:49something that actually works. And all
21:51the prompts and resources from today's
21:53video are in the description, too. Now,
21:55if you want to stop just hearing about
21:57AI and actually start building things
21:59with it, subscribe to Midas. We're
22:01constantly testing free AI tools,
22:03finding alternatives to expensive
22:05software, and sharing practical
22:07workflows that you can use yourself. And
22:09tell me in the comments, which app would
22:11you build first? Thanks for watching,
22:13and I'll see you in the next one.

About this transcript

This page contains the full transcript of Forget Claude Code: DeepSeek + REA Clones Anything 100% FREE (No Paid API) by Maitis, generated from the public captions YouTube serves with the video. The transcript has 3,817 words across 583 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.