The ONLY Setup Guide You'll Need — Claude Code, Codex & Antigravity FREE (Full Tutorial)

Guys, this is going to be the final, ultimate video for those of you

who are dying to use Claude Code, Codex, and Antigravity for free — after this you won't need to search for any other video or trick.

But watch the whole video step by step — if you miss even one part, our setup won't work.

And yes, after the setup I'll also show you a Claude Code login trick that takes it to the next level, so don't miss that.

So let's first show a demo.

This is my Claude Code running in the terminal, and here you can also see the token usage.

Let me also run Codex and show you.

And here you can clearly see I'm using GPT 5.5 at a high level.

Let me give it one prompt: 'Which model are you using?'

And you can see the response too.

That's not all — I'm also using Claude Code with the Desktop app.

Let me clarify one thing: Cowork/Max is only available for paid users, but we're going to use it for free — and I can use Antigravity for free the same way.

Let's see what this video covers.

First we'll set up a smart AI gateway, then connect it with multiple free AI providers, then build a smart combo of all of them, prioritizing the top-tier models.

So if one provider's quota runs out, it automatically switches to another — you won't have to do it manually, our gateway will handle everything.

After that we'll take that endpoint and use it to set up Claude Code in the terminal, Codex, or Antigravity anywhere, then optimize the whole workflow.

That's it — we'll be ready to use Claude Code for free.

So let's start the video.

I also keep posting videos like this on this channel, so subscribe if you want — that's up to you.

That's all the things we need on our system first.

So let's install everything one by one.

Go to this website — I've explained step by step what to install: Node.js, Git, Python, VS Code, plus our Claude Code terminal/desktop, Codex CLI and Antigravity.

So the first step is installing Node.js.

Windows users, copy this command

and paste it into your PowerShell.

Node.js will then get installed automatically on your system and added to the PATH too.

It's already installed on my system, that's why it's showing that.

After that you need to verify it — and here you can see Node.js got installed.

Next, for Mac users: first you'll need to install Homebrew.

Copy this command to install it.

After installing Homebrew you can install Node with 'brew install node', then verify it.

Second step: installing Git.

Windows users, copy this command

and paste it directly into PowerShell.

Git will get installed on your system too — verify it.

For Mac, use 'brew install git', then verify.

After that comes installing Python.

For Windows, this command will install it via PowerShell.

This will automatically install Python on your system and add it to the PATH too, so you won't have to do that yourself.

Then verify it.

For Mac, same process — install Python via brew, then verify.

Next comes VS Code — you can install that yourself.

from its official website — I won't show this step.

Then install whichever AI coding tool you want, like Claude Code — go to its official website.

Based on your system, install the Claude Code Desktop app too.

Then scroll down and you'll get the terminal install command.

Copy it and paste it into your terminal, and Claude Code will get installed there too.

After that, if you want Codex CLI, go to its official website.

Scroll down, copy the 'run in terminal' command and paste it — Codex CLI will get installed.

Then for Antigravity, go to its official website and install it based on your system.

If you want to install Antigravity, copy this command based on your system and paste it in PowerShell.

All the setup is now complete — we're ready to run Claude Code, Codex,

and Antigravity.

The first main part of this video is setting up the AI gateway — for that we're using n8n router, and don't miss a single part, because if you do, the setup won't work.

So let's see how to set up n8n router.

Before setting it up, you need to install n8n router on your system.

Go to this GitHub repo — I'll add the link in my blog description so you can go there directly.

Scroll down, copy the install command, and paste it into PowerShell.

It'll get installed on your system — I already have n8n router installed.

After installing, run it with this command.

Press enter and n8n will open in your browser.

It might first ask for a password — the default one is shown below, use that to log in.

After logging in, go to Providers.

You'll find lots of free providers there that you can connect and use.

Let's connect them one at a time and then create a combo.

First, let's set up Kiro — click Add Connection, then

click AWS Builder ID — it automatically creates your account and connects a free- trial account.

Add as many Kiro accounts as your Gmail accounts — Continue with Google, link a Google account, confirm and continue, allow access.

Our first Kiro account is connected.

Before adding a second account, we need to clear the AWS Builder ID cookies, otherwise it'll just reconnect the same Google account.

For that, go to Settings

then Privacy & Security > Site Data.

You can paste this command directly into your address bar and it'll redirect you to this page.

Then search for AWS.

You'll see the AWS sign-in cookies — delete those, and the amazon.com ones too.

That clears the AWS cookies.

Now let's Add AWS Builder ID again.

Sign up again with a new Google account — select that account, confirm and continue, allow access.

Our second account is added too — add as many accounts as your Google accounts, since each gives free credits.

confirm and continue, allow access — oh sorry, I forgot to clear the cookies.

Let me delete this, refresh and clear AWS cookies again, then Add AWS Builder ID.

It redirects to the sign-up page again — connect with Google, link my Gmail account, confirm and continue, allow access.

Now 3 accounts are added.

Let me show the credits too — each free-trial account gives you 50 free credits.

Account 1: 50 credits.

Account 2: 50 credits (already used some).

Account 3: 50 credits.

And if you have a debit card, you can get the Pro plan for free too.

Let me show that as well — you'd get 1000 free credits.

This is the Kiro Pro plan, 1000 credits a month — click Upgrade.

You'll need to add your debit card — it shows a $0 balance today, so add your card and get the Kiro Pro plan free.

After getting it, cancel the plan

so it doesn't auto-deduct from your card next month.

Go to Settings > Manage Plan, click on the plan, and there'll be a 'Downgrade to Free' option — click it and your credits stay in your account even after downgrading.

That's one way to do it, but I know many don't have a credit or debit card.

So the free account itself gives us 50 credits, which is plenty, since you can add as many accounts as your Gmail accounts.

Our Kiro provider is connected now.

Next let's connect OpenCode.

In OpenCode we'll add models like Big/Pink, DeepSeek, Flash Free — all suggested free models, all available for free.

Let's add all of them

and check whether they work — click the test icon next to each and it'll show automatically if it's working.

Let's check all the models and delete whichever doesn't work.

Almost all won't work since these are free- trial models.

'Internal server error' — this one won't work so let's delete it, since DeepSeek is a good free model anyway, and Big/Pink is good enough too.

OpenCode is set up.

Now let's connect Gemini CLI's free-trial models too — click Add Connection.

Select any of your Gmail accounts, sign in, and get authorized.

All the free-trial models available under that Google account become available.

Let's test which ones work and delete the rest — these ones aren't working.

Gemini 3.1 Flash-Lite preview works, Gemini 2.5 Pro works, and Gemini 2.5 Flash works too.

These four models work — we'll mainly use 2.5 Pro since it's a reasoning model and quite good at it.

To add more accounts to Gemini CLI, click Add.

Say a provider's quota runs out — you can check quota in the quota tracker.

For Gemini Pro, which we activated, you get 1000 credits too.

If that runs out too, add another Google account via Add Connection.

That's it for Gemini CLI.

Now let's add Kiro — click Add Connection.

It'll log in with your account and ask you to sign up.

If it's not set up yet, it'll automatically ask you to.

I've already created an account with an authenticator app, so it's asking for two- step verification.

Let me enter my code.

Verified — the account gets added.

Connect Kiro, generate the key, and it'll be connected.

Our Kiro account is connected now.

Let's check which models we get — I think Mini Maxim 3 will work.

Mini Maxim 3 works.

Let's check K2.7 and K2.6 too, then delete whatever doesn't work.

Only Mini Maxim 3 works, so we'll keep just that.

These are all free providers, but I'll only show the ones with good models.

Now let's go to NVIDIA NIM.

Click Add Connection — first get your API key.

Enter your email

generate a new API key, then copy it.

Come back here, click Add Connection, paste the key, click Check — if it says 'valid', the connection is successful.

Give it a name.

NVIDIA is connected.

Let's check which models work — DeepSeek V4 Pro

is a great model, good at both reasoning and coding.

DeepSeek V4 Flash also works and is a good model too.

So I'll keep only these enabled and delete the rest.

Let's check 5.2 too — I don't think this one will work, so let's delete it and keep just one.

NVIDIA is set up — we'll use this model in our combo.

Next let's set up OpenRouter.

For that we need an API key — create a new key.

Your key gets created.

Go to Add Connection, paste your key, check if it works, give it a name — 'OpenRouter'.

To add accounts in bulk, give a name and your key with a slash.

For now I'm just adding a single account.

Save it, and OpenRouter setup is complete.

We get these free models — let's check which is the best.

Qwen 3 Coder Free is quite good for coding, let's add it, since 'auto' routes to whatever models are available.

Let's check if it's working — 'temporarily rate limited' — okay this one's not working, I'll remove it and check this one instead — OpenRouter Auto — I think it'll work.

This one's working, so we'll keep just this.

Now let's add Gemini too — click Add Connection.

For the API key, go to Google AI Studio.

Click API Key, create a new one.

After creating it, copy the key, come back here, paste it and check — 'valid' means the key works.

Give it a name, say 'Gemini'.

Add another account if you want.

Let's check which model works — Gemini 2.5 Pro.

We'll keep these five models.

If you have GitHub Copilot Pro (student plan), add that too — I have it, so let me add it.

It's your choice.

I won't add it though, since Kiro already covers that.

Let's copy this, paste, continue, authorize — GitHub Copilot Pro is connected too.

It's not working, though

I think we need to add a model — I hadn't added the auto model Copilot Pro/student uses, since we don't manually select models there.

It auto-sends requests to whichever model suits the request — let's add Copilot Chat.

Add Models — this sends requests to quite good models like Claude Haiku 4.5, depending on the request.

So we'll only add Copilot Chat and delete the rest, though you don't have to, since we'll select only this one when building the combo.

That's our GitHub Copilot done.

Next let's add third-party providers too — extra custom providers like Free Models and A0.

They give free trials too, so let's add them — go to Free Models

and create an account there — I explained the whole process in my previous video.

Then go to Usage — each 5-hour window gives you $5 worth of free credit, and you get GPT 5.5 free too

so we'll use GPT 5.5 as well — go to API Keys, then to the OpenAI-compatible format, copy the endpoint, and come to n8n router.

Click 'Add OpenAI Compatible', name it 'Free Model', paste the base URL, create your API key on the site and paste it here, and click Create.

Our API key got created.

Let's check if it's working — 'model not found, try chat validation with model ID' — I think we need to add /v1 at the end, let me check again — now it's valid.

Paste in your API key, model ID is optional but put 'gpt-5.5' anyway since that's what we're using — click Create, then Open, then Add — paste the key

select 'default model: GPT 5.5', and 'Free Model' is connected too.

Let's import from models — every available model gets imported.

We'll use GPT 5.5 — let's test it.

GPT 5.5 works.

There's also Claude Sonnet 4.5 as a workload model, but that one's been banned/suspended right now — to use it you first need to top up, on the 'topup' format.

You'd need to top up first before using it.

But the 'opinion' format point is free to use, and GPT 5.5 is free too, so we'll use that instead.

Also add A0 if you want, since it gives 5 million free tokens daily.

Let me show which models are free there — Mistral Large and Mistral Medium 3.5 are free, and other good models get added from time to time, so keep checking.

For now I don't think we need to add these

since we've already added powerful enough models.

Add it if you want.

There's also Claude4Free, where you get 10 'neurons' per day to use models like Gemini 5.2 and others — add it if you want.

There's one more custom provider, AI/ML API — let's add that too.

It also gives a 5-hour window and lets you use Claude models.

Create an account there — I'll add the link in the description.

You get $140 free and a 2-week free plan.

Go to Dashboard

then to Usage — the 5-hour window is included here too, so let's include this in our setup.

Go to API Keys, create a new key for the router,

copy the secret key, then click Add > OpenAI Compatible, since we're going to use Claude models — name it 'AIML', fix the base URL,

paste your API key here and check if it's working.

'Unexpected response' — let me check again.

Add '/v1' — that should show it as valid.

Once it's valid, click Create, then go to the AIML connection and add your API key,

paste the key you created earlier.

For the default model, leave it as is for now — let's set the default model to Claude Opus 4.8, and keep the rest of the settings as they are for this connection.

Our AIML account is added too, and any available models get imported.

Let's check if they're working — Claude 4 Point works perfectly.

Let's check Claude 5.5 too — it works, so we can use Claude models here.

So all our providers are set up now.

Now comes our main task — creating the combo.

We'll combine all the providers we've added into one combo.

For the combo name, keep it as 'Claude Opus Free'

exactly that, because otherwise Claude Desktop will reject it when we set it up there, since the name won't match — it'll think it's not a genuine Claude model and reject it.

So keep the name exactly like that.

Then click Add Models — we'll set it up with the top 10 models.

First we'll use Claude 5.5 (Sonnet).

Then we'll go to the Kiro models we set up and add Claude Sonnet 4.5 Thinking Plus, since it's best at both coding and thinking.

Then we'll add the free model GPT 5.5, which is also powerful.

After that we'll add Gemini 2.5 Pro — let me find it and add it too, since it's also quite strong at reasoning.

Then we'll add DeepSeek too, since it's also one of the most powerful models — we'll add it from OpenCode: DeepSeek, Flash Free

we'll add that too.

Then, from OpenRouter, the free model — say all these models fail; our fallback will be OpenRouter, which sends requests to whichever free models are available.

After that, if you have GitHub Copilot Pro, add that too

and I'll use it too.

So now we have our full combo of models — first the request goes to Claude 5.5.

Say its quota runs out, meaning the $10, 5-hour window is used up,

then it'll fall back to Kiro's Claude Sonnet 4.5 Thinking (coding), and it'll use up the credits across all the Kiro accounts we've added.

Say one account's 50 credits run out, it'll automatically start using the next account — you don't need to change anything.

Say the Kiro credits run out too, it'll fall back to our free GPT 5.5

for its $10, 5-hour window.

GPT 5.5 will respond.

Say that window runs out too, then it'll go to Gemini 2.5 Pro, and so on for all of them — you can manage the order yourself using these arrows.

But this is the best order to set up, because the top-priority models come first, then the next tier of top models.

That's how the whole setup gets completed — click Create after this.

If you know of any other custom provider, you can add it here too, and it'll automatically get included in the combo.

Now that's done, the next step is setting it up in Claude Code and whatever IDE we're using.

So let's go to CLI Tools.

We're going to set it up in Claude Code — select model, pick the combo we just created.

Keep the default model as 'Claude Free' here too.

Then copy this.

or apply it directly and your settings get configured automatically.

If that doesn't work, you can do it manually — go to your C: drive, then your user folder, then the '.claude' folder, and you'll find a settings.json file.

Open it.

Paste in what you copied from n8n router, save it, then open a new terminal in this folder — I'll open my terminal here — and type 'claude'.

Claude opens up.

Here you can see we're using the 'Claude Opus Free' combo we set up — let's give it a prompt too.

Here's our response — Claude Code responded nicely.

Let's check the logs to see which model actually gave the response.

Let's go to Logs — here it routed to Sonnet 4.5 Thinking Plus, and it checked in — but I think there's an error with AIML, since it didn't respond.

Let's check — okay, found the error: this provider

doesn't support the closed API yet — it doesn't support n8n router.

So we'll use Claude Opus 4.8 instead since it's also a great, powerful model.

Let's go to the combo and edit it — remove the AIML model we set up earlier

and click Add Models, then add Claude Opus 4.8 instead.

It'll be added at the bottom, so use the arrows to move it to the top since our first- priority model should be Claude Opus.

After it, next in line is Claude 5.5, and routing continues accordingly.

Save it, then go to Claude Code and give it a prompt.

Here's our response — checking the log, it says Claude Opus 4.8 gave the response.

Complete — meaning our workflow, the combo we built, is working.

Guys, Claude Code is successfully set up with a powerful model.

And to make it even more powerful, let me show you a plugin you can install in Claude Code called 'Auto Harness'.

In simple terms, Auto Harness gives Claude a self-learning skill.

As you keep working with Claude Code, it silently learns from your sessions in the background.

Whatever you do repeatedly, it turns into a reusable skill on its own, so next time you won't have to re-explain it — it already learned it.

The best part: installing it takes just 30 seconds — go to its GitHub repo

I'll add the link in the description too.

Copy this command and paste it inside Claude Code, then run this command to install it, and that's it — you're ready to use it.

Now let's set this up in the Claude Desktop app too — for that, go to

CLI Tools, then the Claude Desktop app — click the hamburger menu, go to Help > Troubleshooting, and you'll see 'Enable Developer Mode' or something similar there.

Click it to enable developer mode — confirm 'Enable Developer Mode only here'.

Once developer mode is available, you'll see a 'Developer' option, and under that 'Configure Third Party' — click it, and an interface opens.

There you need to enter the gateway base URL.

Copy it from n8n router and paste it into Claude Desktop.

It'll then ask for the gateway API key — enter anything, like I typed 'api-key'.

For the gateway auth scheme, select 'x-api- key'.

and for credential kind, leave it as 'static API key'.

Then click 'Config Changed' and restart it.

Your Claude will restart, and the combo you created will show up — here's 'Claude Opus Free', the combo we created.

Just prompt it with anything — it responded, though right now it says 'failed to start Claude workspace'

which is happening to me because I don't have enough internal storage space — you probably won't see that if you have enough space.

It's still giving a response — 'What can you do for me?' — and here's our response.

So Claude Desktop is set up too, and we can use Cowork here as well for any coding work.

Now let's set this up in Codex CLI too.

Go to CLI Tools, then OpenAI Codex CLI, go to Models, select model.

GPT 5.5, since Codex supports GPT 5.5 best and it makes it more powerful, so we'll select GPT 5.5 here too and click Apply.

It'll automatically write to your config.toml file.

If that doesn't apply, do it manually — go to your C: drive, then the Codex

folder, and you'll find a config.toml file.

You can configure it there directly — set your base URL, and the model to gpt-5.5.

Let's try it out and see if it works.

I'll type 'codex' here.

You can see Codex is running, using GPT 5.5 on GPT-5-Codex.

Let's give it a command: 'What can you do for me?' — and Codex responded too.

So our Claude Code CLI, Claude Desktop, and

Codex CLI, all three are complete.

Next, let's see how to set up Antigravity.

Scroll down and under IDE/MCP Tools you'll find Antigravity — click on it.

To set it up, you need to run n8n router as administrator, so let me run PowerShell as administrator.

First I'll stop n8n router since it's already running,

and then run it again as administrator.

When you start setting up Antigravity, it'll show you a DNS record you need to add to your hosts file — for that,

run this command in PowerShell, and it'll automatically open the hosts file.

Let me open PowerShell for that.

It's open — take the line shown there and paste it in, just as I did when I added mine — leave the rest of the file as is, don't delete anything.

After adding it, you'll get a checkmark showing it's registered successfully.

Then you need to start the server — it needs administrator rights, click 'Start Server'.

The server starts — you can see it running, with a checkmark here too.

After it starts, click on Antigravity — same as before, this is the DNS record we already added at the start.

Then select your model here — say you selected Gemini 3.5 Flash

in Antigravity — which model should the request actually go to?

You can select that here — we'll select our combo for all of them, since that's what we'll be sending requests to.

Setup's done — now let's open Antigravity

if you already had it open, close and reopen it — a chat section will open.

Whatever model you select, your request gets sent directly to the combo we built — say I select Sonnet 4.6 and ask 'which model are you using?' — let me give it a prompt.

And here it says 'I am running on Claude Opus 4.6, how can I help you today' — let's check the request logs.

You can see Opus 4.8 gave our response.

Since we linked it via the Opus name it says that, but the actual request went to our combo — the timing matches too.

So that's how you can set up Antigravity too, and Cline works exactly the same way — set it up there the same way.

After all this, if you want to set it up in any other IDE too,

you can do that too, same as OpenCode — click and apply, and it's set up directly.

Cowork we've already set up.

And for Kiro too,

the process is exactly the same — just select the model, and it'll automatically get set up.

This way, you can do unlimited coding — unlimited Claude Code, Antigravity, and Codex — and set up and check any other free-tier providers too.

And that's it — our setup is complete.

Now you can use Claude Code, Claude Desktop, Codex, and Antigravity through this complete system.

If you've watched all the way to here, congratulations — you now have a permanent setup that people usually keep hunting different videos and tricks for.

If this video helped you even a little, please leave a like, and subscribe to my channel if you want more videos like this.

I'll add all the resources, commands, and documentation used here to my blog, with the link in the description.

If you run into any problem, let me know in the comments and I'll try to help — you can also join my Telegram channel for discussions and updates.

See you in the next video with one more powerful AI workflow — until then, happy coding.