WEBVTT

00:00:00.000 --> 00:00:04.080
In this tutorial, we'll install OpenAI's Codex app on Windows.

00:00:04.860 --> 00:00:11.200
Then, using CC Switch and AI HubMix, we'll make it run on almost any AI model, not just OpenAI's.

00:00:12.720 --> 00:00:15.640
Start in your browser. Search for Codex Download.

00:00:15.640 --> 00:00:31.600
Open the official Codex page on openai.com.

00:00:41.980 --> 00:00:43.700
Click Download for Windows.

00:00:45.640 --> 00:00:56.760
The installer itself is tiny, so it's done in seconds.

00:00:56.760 --> 00:01:02.460
When it finishes, show it in File Explorer.

00:01:02.460 --> 00:01:10.840
Right-click the installer and run it as administrator.

00:01:10.840 --> 00:01:17.920
A setup window opens, and the real download begins.

00:01:17.920 --> 00:01:36.520
This package is about 700 megabytes.

00:01:36.520 --> 00:01:40.040
Fair warning, this part can be slow, usually 5-6 minutes.

00:01:40.820 --> 00:01:41.700
So let's skip the wait.

00:01:49.780 --> 00:01:52.480
And we're back, 96% and counting.

00:02:01.720 --> 00:02:04.880
The moment it hits 100, everything happens on its own.

00:02:06.520 --> 00:02:25.260
The installer closes, and Codex launches itself.

00:02:25.260 --> 00:02:26.900
It lands on a sign-in screen.

00:02:26.900 --> 00:02:28.240
And here's the key part.

00:02:28.240 --> 00:02:29.240
Don't click anything.

00:02:29.240 --> 00:02:31.060
We're never going to log in.

00:02:31.060 --> 00:02:34.280
That's what CC Switch is for.

00:02:34.280 --> 00:02:35.280
Codex is installed.

00:02:35.280 --> 00:02:37.240
Now for the fun part, CC Switch.

00:02:38.020 --> 00:02:42.320
CC Switch is a free, open-source app that can point codecs at third-party providers.

00:02:43.120 --> 00:02:44.980
We'll use it to connect to AI HubMix,

00:02:45.340 --> 00:02:49.900
one gateway that carries basically every AI model behind a single API key.

00:02:50.680 --> 00:02:55.480
Today's example model, DeepSeq V4 Flash, the 0731 release.

00:02:57.080 --> 00:02:59.920
Head to the CC Switch site and hit Download Free.

00:02:59.920 --> 00:03:06.200
It takes us to the downloads page.

00:03:08.280 --> 00:03:09.700
Pick your operating system.

00:03:10.100 --> 00:03:13.180
We're on Windows 10, so we take the X64 installer.

00:03:16.760 --> 00:03:18.980
Chrome flags the file as unverified.

00:03:19.380 --> 00:03:21.200
That's just safe browsing being off.

00:03:21.640 --> 00:03:21.960
Keep it.

00:03:27.400 --> 00:03:28.540
12 megabytes.

00:03:28.820 --> 00:03:29.380
Done instantly.

00:03:29.380 --> 00:03:36.380
Open it to install.

00:03:36.380 --> 00:03:40.380
Setup takes a few seconds, then CC switch is ready.

00:03:40.380 --> 00:03:42.380
Open it up.

00:03:42.380 --> 00:03:45.380
At the top, switch to the OpenAI section.

00:03:45.380 --> 00:03:47.380
That's the one codex reads.

00:03:47.380 --> 00:03:51.380
Then hit the plus to add a provider.

00:03:51.380 --> 00:03:54.380
From the presets, choose AI HubMix.

00:03:54.380 --> 00:03:57.380
Icon and endpoint fill in automatically.

00:03:59.380 --> 00:04:09.560
Next we need an API key, so over to the AI HubMix website.

00:04:09.560 --> 00:04:16.660
Search for iHubMix and open AIHubMix.com.

00:04:16.660 --> 00:04:25.240
One gateway, infinite models, every major LLM behind one unified API.

00:04:25.240 --> 00:04:29.360
Click Console.

00:04:29.360 --> 00:04:41.340
It asks us to sign in Google account, one click.

00:04:41.340 --> 00:04:55.140
Once we're in, open API keys in the sidebar and add a new key.

00:04:55.140 --> 00:05:04.860
Name it DeepSeq V4 Flash 0731 so we know what it's for.

00:05:04.860 --> 00:05:08.820
Notice the options here, quota and an expiry time.

00:05:08.820 --> 00:05:13.620
We'll give this key an expiration date, one day.

00:05:13.620 --> 00:05:15.700
If it ever leaks, it's already dead.

00:05:15.700 --> 00:05:17.900
Cheap insurance.

00:05:17.900 --> 00:05:23.480
Submit, and there's our key.

00:05:23.480 --> 00:05:29.800
it before closing, it's only shown once. Back in CC Switch, paste the key into the AI HubMix provider.

00:05:31.080 --> 00:05:35.960
Bonus feature, endpoint speed testing. Both endpoints answer fast, either one will do.

00:05:39.000 --> 00:05:45.640
Now the default model. Click the download icon and CC Switch pulls the full AI HubMix model list,

00:05:45.640 --> 00:05:47.640
everything they carry.

00:05:49.640 --> 00:05:55.640
Filter for DeepSeq and select DeepSeq V4 Flash 0731.

00:05:57.640 --> 00:05:59.640
Keep the upstream format on responses.

00:05:59.640 --> 00:06:01.640
That's what Codex speaks natively.

00:06:02.640 --> 00:06:04.640
Add it saved.

00:06:05.640 --> 00:06:06.640
The new provider appears at the bottom.

00:06:06.640 --> 00:06:08.640
Click it to enable it.

00:06:08.640 --> 00:06:10.640
Now it's the active config for Codex.

00:06:10.640 --> 00:06:16.640
Time to restart codecs properly.

00:06:16.640 --> 00:06:25.640
Close the window, then quit it from the system tray, bottom right corner.

00:06:25.640 --> 00:06:34.640
And launch it again.

00:06:34.640 --> 00:06:47.000
While it starts, remember last time, it demanded a chat GPT sign in.

00:06:47.000 --> 00:06:49.280
This time, straight to the welcome screen.

00:06:49.280 --> 00:06:50.500
No login at all.

00:06:50.500 --> 00:06:56.320
CC switch already wired codecs to AI HubMix.

00:06:56.320 --> 00:07:04.220
Pick any field, it doesn't matter, and continue.

00:07:04.220 --> 00:07:05.640
It offers a demo task.

00:07:05.640 --> 00:07:07.640
Skip it.

00:07:07.640 --> 00:07:10.220
Confirm.

00:07:10.220 --> 00:07:14.980
Go to Chat GPT.

00:07:14.980 --> 00:07:15.980
One last pop-up.

00:07:15.980 --> 00:07:18.520
Pitching GPT 5.6.

00:07:18.520 --> 00:07:20.140
Take Continue with Current Model.

00:07:20.140 --> 00:07:22.020
So we stay on our custom setup.

00:07:22.020 --> 00:07:23.400
Here's the main screen.

00:07:23.400 --> 00:07:24.860
Bottom right of the input box.

00:07:24.860 --> 00:07:26.620
The model says Custom.

00:07:26.620 --> 00:07:30.460
That's our DeepSeq V4 Flash, riding through AI HubMix.

00:07:30.460 --> 00:07:31.820
Time for a real test.

00:07:31.820 --> 00:07:44.820
I'll ask for a short introduction of AI HubMix.

00:07:44.820 --> 00:07:47.373
Questions in, send it.

00:07:48.893 --> 00:07:53.173
The request flows through AI HubMix to DeepSeq and Codex gets to work.

00:07:53.173 --> 00:07:57.613
It's an agent, so instead of guessing, it actually runs a shell command to look around

00:07:57.613 --> 00:07:58.613
first.

00:08:02.373 --> 00:08:05.473
21 seconds of work.

00:08:07.753 --> 00:08:08.893
And here's the reply.

00:08:10.953 --> 00:08:11.993
It's an honest one.

00:08:12.373 --> 00:08:15.533
The model says it doesn't have reliable information about AI HubMix

00:08:15.533 --> 00:08:19.293
and asks for a URL or some context instead of making things up.

00:08:20.033 --> 00:08:20.633
Fine by us.

00:08:21.033 --> 00:08:22.413
Next, we'll hand it the URL.

00:08:24.333 --> 00:08:27.313
Picking up right where we left off, the model wanted a URL.

00:08:27.753 --> 00:08:28.353
Let's give it one.

00:08:28.353 --> 00:08:35.353
The answer, an AI model aggregator, and the URL is AIHubMix.com.

00:08:40.213 --> 00:08:41.873
Sent. Now watch it work.

00:08:45.373 --> 00:08:50.433
It announces the plan, fetch info from the site, and fires off Curl to download the homepage.

00:08:50.433 --> 00:08:57.833
The first attempt hits a Windows quirk.

00:08:57.833 --> 00:09:01.793
In PowerShell, curl is just an alias for InvokeWebRequest.

00:09:01.793 --> 00:09:09.933
The model catches that and switches to the real curl.exe.

00:09:09.933 --> 00:09:20.413
Still blocked, so it retries, this time pretending to be a browser with a Mozilla user agent.

00:09:20.413 --> 00:09:22.413
This is what makes agents fun to watch.

00:09:22.413 --> 00:09:26.413
It's debugging its own path to the answer.

00:09:30.413 --> 00:09:38.413
And remember, this is DeepSeq v4 Flash, a fast, inexpensive model doing genuine multi-step work.

00:09:44.413 --> 00:09:49.413
Once the page downloads, it strips the HTML down to readable text.

00:09:50.413 --> 00:09:59.473
1 minute, 31 seconds of work.

00:10:01.333 --> 00:10:02.553
And here's the payoff.

00:10:04.153 --> 00:10:06.853
A real introduction, sourced from the live site.

00:10:07.653 --> 00:10:13.513
One API, many models, and open AI-compatible endpoint, routing, developer tools, the works.

00:10:14.213 --> 00:10:14.753
Final stop.

00:10:15.133 --> 00:10:16.333
Let's see what all of this costs.

00:10:16.733 --> 00:10:18.293
Back to the AI HubMix console.

00:10:18.853 --> 00:10:20.333
In the sidebar, open logs.

00:10:20.413 --> 00:10:24.093
Here is every request itemized.

00:10:24.093 --> 00:10:32.593
Each row shows the model, our deep seek, v4, flash, 0731, tokens in and out, cash tokens,

00:10:32.593 --> 00:10:36.193
latency around a second and a half, and the exact spend.

00:10:36.193 --> 00:10:39.213
Most of these calls cost less than a tenth of a cent.

00:10:39.213 --> 00:10:43.913
And that's the full setup, codecs on windows, CC switch in the middle, and one AI Hub Mix

00:10:43.913 --> 00:10:46.733
key that unlocks any model you want to try.

00:10:46.733 --> 00:10:47.413
Thanks for watching.