It was only a week ago that I was saying

codec cli wasn't quite as good as claude

code though GPT5 was a really good model

blah blah blah all of this noise about

it's not as good it's hard to use it's

kind of hard on the eyes can't figure

out what it's saying. Fast forward 2

days after I released that video and

they dropped such a big release and so

meaningful that I just had to create

another video to catch up with what they

did in only one week. They really

covered just about everything that I was

complaining about, at least partially in

many, many of those cases. I want to

dive into that so that you can see all

of the moving parts and I'll try to

contextualize each one so that you

understand if you're bumping into these

or if you've previously bumped into

them, maybe some of them are entirely

gone. So, we're going to dive in and

take a look at all of these tiny little

changes that they made that amount to a

major major update. So, first let's take

a look at what Codeex CLI is for the

people that aren't aware. I will not

cover this very deeply. Check out other

videos, but I'll give you an extremely

brief overview of what we're looking at

to describe what this compares to. But

we got a lot to do. Let's dive in. Cloud

Code is kind of a terminalbased code

editor that people are using. Codeex is

the OpenAI version of Claude Code. So,

if you've seen Claude Code before,

that's what Codeex is. Number one, if

you've seen cursor or maybe even

something like GitHub copilot, those

kinds of things, that's really what

we're talking about. So that's all codec

is. We're about to dive in to some of

the features that are inside of this

thing and how cool they are. So if you

don't know, kind of now you know. I'm

going to be in cursor and I'm going to

pull open a panel that will maybe look a

little bit like the clawed code panel.

In fact, I will use the same keystroke

to open cloud code and I will stop the

execution of cloud code and just get

codeex running here. So that's all I'm

doing. It says claude code up top. I'll

rename this thing, but the one thing I

can't do is change the icon. So maybe I

can, but it doesn't matter. You know

what I'm doing. All right, so let's step

into some of the changes that we've seen

here in in Codeex CLI and really address

some of the things that I was

complaining about. Okay, number one,

cognitive cost. This was the problem

that I had last time. Big deal. As you

were using codecs, this was really a

week ago. Um, as you were using codecs,

everything that GPT5 was thinking was in

the clear, so you could see it. So, if I

just kind of say something like, tell me

about the code, then it starts working.

And you can see it actually is already

kind of reusing some of its surface

area. And very quickly, you'll see that

it is starting to summarize its thinking

as it looks at each one of these files

instead of just spitting it all out. All

right, so here it is kind of giving me

the report. I will say this one's still

a gap with Claude Code. Cloud code does

a much better job of kind of the styling

of what's occurring when it prints out

the information coming back. It kind of

treats more things like markdown. You

can see it's getting much better. If you

remember a week ago, we would have two

or three hashes in front of this or the

stars making it bold or something like

that. So, they've definitely gotten

better, but they haven't turned things

into an organized structured markdown

look yet. I think they'll get there, but

this is so much better already. And all

of this information up here was

collapsed. I didn't have to see all of

the thinking. In fact, that leads us to

a question. What happens if you want to

see the thinking? This is the second

thing that they've covered. You can see

down here they actually have this new

call out for controlt to show the

transcript. So if I hit control T then

we start scrolling through all of this

and you can see there's the file that it

gave back and then it's you can see the

thinking that we didn't see before. I'm

considering potential pitfalls or

constraints like no drag classes. I

might include file paths. So the the

entire thinking here is shown and in

fact many of the the files that we would

be looking at of course are shown here

as well. And this was really a lot of

this was not all of it but a lot of it

was kind of coming out in the console

previously which made it incredibly

difficult to understand when it was

talking to us. All right, how do you get

out of this? Q gets you back to exactly

where you were. So at any time you can

kind of jump in and out of the

transcript including while it's working.

So really great job here from a terminal

standpoint. I think this gets us very

close to what Cloud Code is doing which

just happens to be best in class right

now. I will say that another thing that

they did very briefly is they worked.

I'm now using my scroll wheel instead of

my arrow keys to move through. So, they

also included some changes to kind of

quality of life moving around in the

system. If you were tired of having to

use your arrow keys to move, they got

you. All right. Here's another cool one

that I think is really useful. I'm not

sure it's necessarily important. Uh some

people find it more important than

others. So, let's imagine we say, "Tell

me about the coverage here." and it goes

off and starts reading files, checking

out all of the tests, those kinds of

things. I have some are system test,

some are into tests and I can say not

the E to E tests and now I can enter

that while it's working. So they have

created kind of a a queue of messages

that can be put in. Now it's important

to understand I think a big distinction

here. It feels a lot more like this is

the next action message. Cloud code has

a pattern when you kind of inject these

thoughts that it'll pick it up on the

fly. It'll get through some of its steps

and as it comes back to its agentic loop

at some point it picks up the most

recent user messages replans from there.

That doesn't seem to be occurring here

in Codex. I feel like what Codex is

actually doing is it's given us just

simply a queue of messages that we can

keep queuing up. And you can see it

finished. It gave me the sound effect

saying, "Yep, I'm done." and then it

started working on the next task. But

this would allow you to be able to type

and think out loud and do some things

while it's working. So, it's better than

nothing. It's getting there. And they

will definitely cover actually

interjecting the message into the

thought process as it's working in at

some point. So, I'm sure that's coming.

All right. Another one that was really

really major. If you recalled, we would

have to drop out of this, go either

start with a new command or go into our

TOML file, just the configuration system

in your home directory to make some

changes on how your system started and

be able to start it with different

models. That was the way that you were

able to basically tune how much thinking

or which model you wanted to use. Well,

no longer. They also addressed that with

a slash command of model. So now you can

see the different models that you can

use and select any one of them. minimal,

low, medium, or high at any time. So, if

I go into high mode now, all of my

interactions are on high. Now, I will

say Cloud Code has since released

something. This one was a later release

as well, that they have a status line

that you can leave, what model you're

using, and other information on the

status line. I really would expect

codecs to go there as well because it'd

be nice to know what model I'm using at

any given moment instead of having to

try to remember that or go back into

info or status so that you can come in

here and see what model you're working

on. Okay. And the kind of the last one

in this area of current cognitive pains.

There was no good way to resume

previously to kind of go back to the

previous conversation that you were in.

They have supposedly added that. I am

having problems getting to it.

Theoretically, the way that the docs

read is you should be able to hit escape

twice and be able to see a list of your

previous conversations, select one, and

be in that context. Today, if I hit

escape twice, what I end up in is my

transcript. Essentially, we saw this

previously. This is just what has been

happening entire in the entire thought

chain essentially. And I don't see a way

in here for me to be able to jump into

this point in the context or something

else. So, I think maybe there's a defect

in here or something. I don't

understand. Please leave me a comment if

I don't get it. I'm missing this. I

looked. I'd love to be able to do it.

They have added a note saying they put

it in. So, obviously, this is coming

very soon. Okay. And another one that I

really like that I think is a big shift

and is going to help quite a few people.

I'll bring this tool. This is what this

repo is all about. In fact, it's a timer

that I wrote that I use when I record

these videos that helps. So, if I took a

screenshot of this, you'll see my

screenshot utility fly out here, I can

drop this in, and you can see that it's

dropping the image in instead of some

big file reference to some path that

this might not have reference to be able

to read. That was a pain before. You

might have had to bring the actual file

into the folder structure that you're in

and then pull it in. Kind of a big pain

in the butt, but okay. Here's the next

thing I'll show you that y'all have seen

before. I use a clipboard system and on

my clipboard is that screenshot. So I

can also if I just controlV now it's

important for me to say this out loud

for people on a Mac it's controlv I know

Windows people are saying whatever it's

been controlv for decades it's commandv

typically on a Mac to paste but when

you're in a terminal application control

is a very common sequence character

that's used so it's controlv to paste

here cloud code has also added command v

support they had controlv first they now

have both so I think codeex will also

cap catch up in this space but this is a

massive win. If you have screenshots and

you're trying to figure out how to

elegantly get them in here, this

absolutely was a big ding for me last

time. This one would keep me from using

the product religiously. It's fixed.

Okay, here's a little odd one. I'm going

to use a very simple example here that

absolutely would need web search. What

they've added is a web search tool

inside of the system. This gets a little

bit strange because actually if you ask

this question or things like this

sometimes it will go into a curl and

write its own tooling to be able to

access the internet in some ways but

this is one of the things they have an

actual web search tool that can be

enabled. Ah yes so here we go. So you

can see that it's writing curl to try to

answer this question figured out where

it might go to get the information just

happens to know that from training. So

it may very well not have the right API

etc etc. However, if you run codecs with

a d- search parameter, then you have the

actual search feature in here. And I

just use the up arrow key. Another

quality of life change. Uh, if you go

through the up and down arrow keys,

it'll go through your previous user

messages. Kind of a great extra feature

here. And I'm going to submit this one

once again. Planning weather tool call.

Searched for weather in San Francisco.

So, that's great. You can see that it

actually used the internet in this case

was far faster. So, if you use search,

it's already unlocked for you. Okay,

here's a major one. This one's really

cool, though I don't think it's done

yet. It's definitely definitely not done

yet. All right, so what they've added,

you know, these slash commands when you

hit slash, you have the ability to find

one of these commands and be able to run

it yourself. Those are slash commands

that of course help you just run an

action, but they are backed by prompt

very often. And so, one of the things

that the Codeex team has released is

something they're calling a prompt

library. So the way to do this is go to

your home directory. I am just going to

open up my codeex folder which happens

to be in my home directory in cursor

here so that you can see the files that

are inside of it. But it's just

basically your home folder and a codec

folder inside of it. Inside of that are

all of these things or you'll have a

varying mix of these things in there. I

have created a new folder called prompts

and any markdown file that you put in

here it will use as a custom prompt. I

have given us an example one here. I

have named it joke because the name of

the file here is what this the slash

command actually is used as. And what

you're seeing here is it tell says tell

me a joke. If the user gives you context

in any way, use that context to help

form the joke. Otherwise, make a joke as

if telling an 8-year-old a joke as a

grandfather. Okay, so that's the prompt.

Let's go give this thing a shot. So, if

I do dot slash joke here, you'll see

that we have slash joke. All right, I

hit enter. You'll see that it plops the

entire context down so that you can see

what prompt was stored in your prompt

library. It just drops it in here. And

you'll see grandpa's voice. Why is our

electron timer such a good babysitter?

Because it's always on top. Keeps an eye

on OBS. And if the connection wanders

off, it comes back every 10 seconds for

a websocket hug. All right. What I will

say here is that the thing that they

don't have right is these are are

nondamic. So, weirdly enough, they've

left them as just a static prompt

library. And maybe this is just V1 of

the release. So, if I said, let's say,

joke about dogs, make it easy. You'll

see that the only thing that shows up is

the context that comes from this file.

This entire file replaces anything I

type in. So, it won't come, you can't

pass any context into it. So, it's

non-dynamic. It is a static thing.

Whatever's in this file flops in no

matter what you want to say. That's

really still a quality of life problem.

I would expect that they're really

heavily working on this and getting a

lot of feedback, but still a great step

that you can make slash commands. Oh,

yeah. So, here's the problem. They're

just prompts. First off, I said that and

I want to make sure that that's very

clear. They're also up in your root

folder, right? This is back in your home

folder. So, it's basically userle

prompts. You can't do it at a project

level, but I bet they're working on it.

I bet they're hearing it. Okay. All

right. So, we've seen all of this

terminal stuff. Let's get out of this

terminal thing. It's kind of a pain,

right? It's still terminal. It still

doesn't format very nicely. It would be

nice to be able to see this a little bit

better. One of the things that you can

do in these systems is go into the

extensions. And in extensions, if you

search for codecs, you'll see that there

is now an extension specifically for

codecs in all VS Code environments. If

you install this, what you end up

getting is a little button like this.

Now, I am in cursor, so the buttons are

across the top. That's because we're in

a cursor environment. That's not the

case for VS Code. And I am going to want

to move this to the other side. I'll

show you why in a second. Okay, this

seems like a reasonable workflow. I'm

going to then open codeex. And hey, by

the way, welcome to codeex in cursor

itself. This is not cursor. This is not

the cursor panel. Remember, this is the

codeex panel. That little label becomes

important in just a second. But as I ask

questions here, we get to see the files.

That's great. But really the reason that

I actually need to use something like

cursor, VS Code, or some other IDE to

wrap these systems is so that I can see

the files that I'm talking about while

I'm communicating with the agent. So

having these in the same panel, that's a

real problem. So what I really need to

do is I need to move this over to this

panel. And this, by the way, is cursor.

Don't get confused. I understand it's

we're really in inception at this

moment. And it's going to say new chat.

So, I'm just going to get you used to

the fact that cursor says new chat over

here. What we want to do is drag this

object and drag it over here. The only

way to drag this object. I know it's

involved. So, we need to go to settings

and you need to search for activity bar.

This is the activity bar and the

activity bar in cursor. Now, like I

said, if you're in VS Code, it's already

vertical. In cursor, they've made it

horizontal, so it collapses up at the

top here. You will want to make that

vertical. And we will need to restart.

All right. And now we're back. And over

here you can see here is the codeex

panel. Here is where we were previously.

Now I have an object that I can drag

around. This is how you drag object. And

now I can drag this object over here

inside of the codeex or inside of the

cursor panel itself. I'll move it to the

front so that it's basically the first

thing that's selected when we come into

this panel. And this new chat, this is

what I was telling you. That's cursor

back there. If you need to get to

cursor, I know this is really messy.

They'll fix this. This will get better.

But in any case, I'm going to leave it

on this side so that we can see both the

conversation that we're having as well

as the files because that's a much more

logical way to set this up. Okay, I've

been using this panel of course now for

a day and a half because it's brand new.

It is fantastic. They've done some

really wonderful work here. They really

nailed just about everything here. I

will take you through the brief brief

highstrokes here. Down at the bottom is

a very exciting set of controls, right?

This is the thing that we just saw with

/model. This is basically that slash

command. You have different kinds of

conversational levels so that you can

set it into a chat mode which just means

it won't write files or an agent mode

which is the standard mode. I have mine

in agent full access so that I don't get

all of these prompts that will come back

and say do you want me to run this

command? Those kinds of things. But

there's also you can run things in the

cloud or locally because you know codec

CLI has this ability to kind of also be

a cloud-based system. So you can attach

different projects to your cloud

environment and run them that way. The

settings, this would be the place some

of y'all might be asking, hey, does this

support MCPS immediately? Yes, it does.

This is the place that you'll come in

and read about it and kind of the config

tumml that will be up in your home

directory kind of stuff. That's where

you'll be messing with stuff. But okay,

when you want to create a new chat, this

is the button that you'll hit to create

a new chat. Basically saying clear my

context and start over. So let's pose a

question to them. Tell me about the

coverage. Take a look around and tell me

about gaps. And this is what it looks

like when it starts to fill in. There's

a couple things that I'll point out

here. Number one, they are collapsing.

So, all of these are different collapsed

sections. You can come into each one of

them and if you look inside of it, you

can see both chain of thought as well as

as some of the contents of the files and

other information in each one of these

steps. What I want to call out here

that's not obvious, I I missed it for

quite a while. This is a very excellent

panel. It just it sits right above the

text area where you're entering your

request. And this is its to-do list. So,

if you if you're using Cloud Code, you

know, Cloud Code has a very strong

backing with its to-do list. Does a

great job. But in any case, this is a

great way to watch what step it's

working on and where it is and can be

collapsed out of the way at any time. I

really wanted to call this out because I

didn't see it in the beginning and I

thought that's really missing that I

don't know how much more it has to do

cuz it just sits down there and says one

out of four tasks and it just never

struck me. But that panel I think is a

lovely panel. All right. And you

remember when I was complaining that

back in the CLI world, it was not

organizing things correctly. We

basically ran the same kind of request

here to give us a report of the

information of coverage. And you can see

now it looks a lot more like a standard

kind of markdown report that you might

get. Makes it a little bit easier to

read. Really, I will say it's so

verbose. A lot of these things,

especially when you ask for something

like a report, we still need to solve

this puzzle. I wish someone would figure

out a way to say the information in here

very much like a canvas can be dropped

into a file in something like this or as

some temporary file or something that

allows me to read it better to store it

in some location allows me to easily

save it and not have to use copy and

paste mechanisms to try to move it over

which by the way I will also say is a

shortcoming here or as far as I've been

able to tell is a much bigger

shortcoming that I can't find any easy

copy uh kind of disclosure exposures. Uh

cursor itself, if you did something like

this, will give you copy buttons all

over the place. Chat GPT gives you copy

buttons. They don't have those here.

Little bit of a shortcoming, but really,

man, this thing is awesome. It is

constrained. It's a little bit smaller.

You can stretch it out and kind of

collapse other things if you like to to

really get a full view of it. I think

I'm a convert. I really want to start

using this. So, I'm going to daily drive

this for a while. All right. All right.

That was a lot. Let's let's shut this

down. Okay. So, I really felt like I

needed to call this stuff back out

because I didn't mean to give them a

black eye last time, but I think some of

the things that I was saying, though I

believe they were fair, they were

criticisms and a little bit little bit

direct. And this addressed so many of

them 2 days later that I was just

astonished. Obviously, they've been

working for a really long time and I

cannot advise them enough. They've had

three releases since that. cannot advise

enough that they need to just continue

to release in tiny increments like this

because that's what we need. That's what

we want to see. That's how we're going

to gain this adoption and this feeling

that this project is ever moving and

ever updating. Every time I put a a

request into Cloud Code, the next thing

that comes down is, hey, there's a new

Claude Code instance. Go install the new

version. So, I get a sense that, boy,

this is actively being improved. I think

the same thing needs to be done here.

And by the way, this week you guys just

completely nailed it. This is a a a out

ofthe-park kind of home run thing that

you just hit. So great freaking work. I

really wanted to call that out. Uh I'm

really glad everybody was here. I hope

this helped a little bit. And maybe if

you had some concerns, maybe this is the

one that kind of made you go, "Oh, wait

a second. That kind of feels like parody

now. It's pretty darn close." All right.

Thanks for coming along for the ride on

this one and I'll see you in the next