In this tutorial, we'll install OpenAI's Codex app on Windows.
Then, using CC Switch and AI HubMix, we'll make it run on almost any AI model, not just OpenAI's.
Start in your browser. Search for Codex Download.
Open the official Codex page on openai.com.
Click Download for Windows.
The installer itself is tiny, so it's done in seconds.
When it finishes, show it in File Explorer.
Right-click the installer and run it as administrator.
A setup window opens, and the real download begins.
This package is about 700 megabytes.
Fair warning, this part can be slow, usually 5-6 minutes.
So let's skip the wait.
And we're back, 96% and counting.
The moment it hits 100, everything happens on its own.
The installer closes, and Codex launches itself.
It lands on a sign-in screen.
And here's the key part.
Don't click anything.
We're never going to log in.
That's what CC Switch is for.
Codex is installed.
Now for the fun part, CC Switch.
CC Switch is a free, open-source app that can point codecs at third-party providers.
We'll use it to connect to AI HubMix,
one gateway that carries basically every AI model behind a single API key.
Today's example model, DeepSeq V4 Flash, the 0731 release.
Head to the CC Switch site and hit Download Free.
It takes us to the downloads page.
Pick your operating system.
We're on Windows 10, so we take the X64 installer.
Chrome flags the file as unverified.
That's just safe browsing being off.
Keep it.
12 megabytes.
Done instantly.
Open it to install.
Setup takes a few seconds, then CC switch is ready.
Open it up.
At the top, switch to the OpenAI section.
That's the one codex reads.
Then hit the plus to add a provider.
From the presets, choose AI HubMix.
Icon and endpoint fill in automatically.
Next we need an API key, so over to the AI HubMix website.
Search for iHubMix and open AIHubMix.com.
One gateway, infinite models, every major LLM behind one unified API.
Click Console.
It asks us to sign in Google account, one click.
Once we're in, open API keys in the sidebar and add a new key.
Name it DeepSeq V4 Flash 0731 so we know what it's for.
Notice the options here, quota and an expiry time.
We'll give this key an expiration date, one day.
If it ever leaks, it's already dead.
Cheap insurance.
Submit, and there's our key.
it before closing, it's only shown once. Back in CC Switch, paste the key into the AI HubMix provider.
Bonus feature, endpoint speed testing. Both endpoints answer fast, either one will do.
Now the default model. Click the download icon and CC Switch pulls the full AI HubMix model list,
everything they carry.
Filter for DeepSeq and select DeepSeq V4 Flash 0731.
Keep the upstream format on responses.
That's what Codex speaks natively.
Add it saved.
The new provider appears at the bottom.
Click it to enable it.
Now it's the active config for Codex.
Time to restart codecs properly.
Close the window, then quit it from the system tray, bottom right corner.
And launch it again.
While it starts, remember last time, it demanded a chat GPT sign in.
This time, straight to the welcome screen.
No login at all.
CC switch already wired codecs to AI HubMix.
Pick any field, it doesn't matter, and continue.
It offers a demo task.
Skip it.
Confirm.
Go to Chat GPT.
One last pop-up.
Pitching GPT 5.6.
Take Continue with Current Model.
So we stay on our custom setup.
Here's the main screen.
Bottom right of the input box.
The model says Custom.
That's our DeepSeq V4 Flash, riding through AI HubMix.
Time for a real test.
I'll ask for a short introduction of AI HubMix.
Questions in, send it.
The request flows through AI HubMix to DeepSeq and Codex gets to work.
It's an agent, so instead of guessing, it actually runs a shell command to look around
first.
21 seconds of work.
And here's the reply.
It's an honest one.
The model says it doesn't have reliable information about AI HubMix
and asks for a URL or some context instead of making things up.
Fine by us.
Next, we'll hand it the URL.
Picking up right where we left off, the model wanted a URL.
Let's give it one.
The answer, an AI model aggregator, and the URL is AIHubMix.com.
Sent. Now watch it work.
It announces the plan, fetch info from the site, and fires off Curl to download the homepage.
The first attempt hits a Windows quirk.
In PowerShell, curl is just an alias for InvokeWebRequest.
The model catches that and switches to the real curl.exe.
Still blocked, so it retries, this time pretending to be a browser with a Mozilla user agent.
This is what makes agents fun to watch.
It's debugging its own path to the answer.
And remember, this is DeepSeq v4 Flash, a fast, inexpensive model doing genuine multi-step work.
Once the page downloads, it strips the HTML down to readable text.
1 minute, 31 seconds of work.
And here's the payoff.
A real introduction, sourced from the live site.
One API, many models, and open AI-compatible endpoint, routing, developer tools, the works.
Final stop.
Let's see what all of this costs.
Back to the AI HubMix console.
In the sidebar, open logs.
Here is every request itemized.
Each row shows the model, our deep seek, v4, flash, 0731, tokens in and out, cash tokens,
latency around a second and a half, and the exact spend.
Most of these calls cost less than a tenth of a cent.
And that's the full setup, codecs on windows, CC switch in the middle, and one AI Hub Mix
key that unlocks any model you want to try.
Thanks for watching.