Oh My OpenCode Is Actually Insane

Have you ever wondered if you can run agents in parallel and have some agents even work in the background while you do AI coding?

Well, if so, then this video is for you.

We're going to be talking about a plugin called My Open Code.

And essentially, what is My Open Code?

It is one plugin coordinated with seven specialized agents to do what one agent can't.

And that's what My Open Code is.

And the results are honestly surprising.

So let's just jump into what is My Open Code, how you can install it, and at the end, how it actually works under the hood for all the nodes watching.

Anyway, let's jump straight to the video.

All right, so get in.

We're looking at My Open Code as stated in the intro.

And one thing I do want to say off the get go, you can actually use different subscriptions here.

You can actually use your Claude, your Gemini, and your chat GPT subscriptions under the hood.

And an external plugin that they use to kind of connect to those servers so they can actually use in my open code.

So if you do have multiple subscriptions, you can definitely use this.

And I would say it's probably best because this really does eat tokens very quickly.

To install my open code, you got two options.

You got the human option and you've got the LM option to do it.

Like, so if the human option, you can just run this command and install it.

One of the prompts to install the different chat descriptions, or you can follow the description here.

So I'm actually going to do the LM side of this one because I think it's kind of interesting to see.

So I'm literally going to open open code on my my local.

So at the moment, this is have a look.

We've got the build and we've got the plan agent.

All right, so I'm going to run open code on my local here.

You can see I've got nothing in this folder at the moment.

I'm just going to jump straight back to here.

I'm actually going to grab this file here and we're just going to tell it to install.

So I'm just going to come back here and I'm going to say, please install.

And you can even use any model to open code here.

And we can see it will start it'll grip that and start running through.

So this just makes this a bit bigger.

And you can see it's asking if I have the Claude Max Pro subscription.

I don't have the 20 max.

I just have the standard.

So I'm going to say one standard and then two.

Do you have a chat GPD description?

I'm going to say yes.

And do you want to integrate gem nine models?

I'm going to do say three.

No, well, for this case.

And it'll use this and then help install the rest of it.

I think this is kind of a neat way of just installing it, you know, using.

So one thing I am going to do here, I'm going to stop this.

So it will install into my global.

I actually want to install my own project.

So I'm going to just say, hey, let's call this.

So I'm going to tell this to only install on the project level only.

The reason being is I just don't want this in my global because that means every project that I open, this will be in.

And one thing to note with the plugin, it runs by default.

So even if I'm not using the agents, the hooks are still going to happen.

I just don't want that.

So I'm just going to make sure it installs on the local.

So we can see here, it's just setting up everything locally for us, which is great.

So everything is set up.

We'll see here, it is telling us a few things that we need to do.

We need to open code or log in like provider, Claude max, follow that within traffic and for the chat GPT, we need to go to open code, Jason, do the plugin and make sure we've got the Codex or plugin enabled so we can actually use our Codex.

Point.

So let's just quickly run through this.

All right.

So let's just do this one together.

So everyone knows how this works.

So I'm just going to go to a new terminal.

I'm just going to go open code.

Or and then I'm just going to set a login.

This will allow you to do for any provider.

So I'm going to go in traffic here and I'll record max.

And from here, I already have a Claude max.

So I'd literally just press enter here.

It will then prompt me to go to Claude and say, authorize for the token.

And then it will be linked.

So this is already linked for me.

So I'm not going to go through all of this, but you will follow that along.

All right.

And then this is go lastly to the next one that it stated.

It said we're going to make sure that this plugin is installed.

So if we go here and we go packages, see if the plugin is installed.

Nope.

I do believe this plugin is installed on my global.

I'm just going to say, can you install the in the local project?

So this makes it easier.

So I don't really have to grab it all, but it should just install it.

So we should see it being popped up here as a dependency.

All right.

So we can see it's being added.

All right.

So we can see it added the plugin and plugins there.

And all we need to do now is open code auth login.

We'll do the provider open AI and then we'll choose our subscription.

All right.

So just jumping through to the other terminal again, I'm just going to go open code auth and then login as stated.

And I'm literally going to go to open AI.

It's going to ask for the code of subscription.

It's going to take me to the page.

I'm going to log in and with that, I will have my codecs enabled.

I'm not going to show you the rest of it, but I'm just going to go through the process now.

All right.

At the end, you should see a screen like this.

Just codecs enabled.

Great.

So we jump back to the terminal.

We can see login was successful.

Perfect.

Cool.

All right.

So we've kind of done the flow.

This is quickly jump into what it was installed here.

Just say everyone knows.

So first off, we can see the plugin has been installed in the packages.

So we got this dependency open code AI plugin, and then we've got open code AI codecs.

And if we go to the open code JSON here, we will see it is pointing to our plugin, OMA open code.

In case everyone doesn't know how this works, this does an MPM install on the global.

So it'll just grab OMA open code at every time.

So what will happen is this will update consistently.

When OMA open code gets an update, you will get an automatic update.

You don't really have to worry about fetching it if you don't want that.

And if we go to OMA open code Jason here, we can see these different ones that Oracle is using the check GPT 5.2.

This is like a search.

We got a librarian and then we've got our explore model, which is just using grok code.

This is still like explore the code base.

That's the main one.

All right.

How does my open code work under the hood?

Well, it is a plugin.

And just on screen here, you'll see we've got OMA open code right here, which is a CLI.

And OMA open code literally replaces the plan and build agent.

So this will replace the plan and this will replace the bolt.

One thing I do want to stipulate here is it does use the Opus 4.5 with extended thinking by default, which can be kind of expensive if you don't have the maxed out plan times 20.

So you keep that in mind, you might want to change that setting.

But essentially you have the seven agents that work with each other to perform certain tasks to get more done.

All right.

I do want to state that the document agent and the front end engineer can run in the background.

So it doesn't always need to work with everything.

So those can run in the background while other tasks are completing, which is kind of neat so you can complete things faster.

Also, you'll notice that it has LSP diagnostics built in and it has its own MCP service that they can use to explore, you know, call different services and get better results.

We'll jump into what each of these are in a bit.

But essentially this is like the main flow of what's happening.

So you get your one orchestration agent here that then delegates and works with other sub-agents to measure and make sure the output matches what you expect.

And in a nutshell what it does, but it will jump into more detail how they get running with it.

All right.

This is another just quick look at the diagram, just showing you how it all works again.

Again, we've got the user input, we've got open code CLI, and we've got the open code, oh my open code plugin.

And we have a few different things.

One is the hook system.

So it's got 21 lifecycle hooks that are built in with some good tools that are also associated to it, which is that LSP tools that I mentioned before, 11 of them.

It's got the grip tool, session management, background tasks, interactive bash.

And then it also has access to the MCP system that I was talking about, some of them being my favorite like context seven.

We've got grip-app or GitHub search.

This is really useful if you want to reference other external projects, and you want to like understand how they were building their code base, you can mimic that using this.

I can grab that information quite neatly.

And you can also use web search extra, which just does a simple web search for you.

You might reach the limit and you might have to buy your own token, but very powerful indeed.

And then you've got the agent system that I talked about earlier, that I won't go into too much detail on this one.

Anyway, so that's my open code at a quick glance.

This is jumping to show you how it works.

All right, so now we need to rerun open code to get this new affection.

So I'm going to make this big and I'm just going to exit this and I'm going to run open code again.

And we will see now that the build agent has changed to OMA open codes.

So first, and we can see OMA open code is working because the build agent has changed.

And if I press tab now, you'll see the plan agent has also changed to OMA open codes one.

So one thing to note is you can see it is using Opus 4.5 latest.

Remember, this is with extra thinking.

So it will consume your resources quite quickly.

I'm actually going to just change this.

Yeah, because I did an experiment with this the other day and it destroyed my tokens within two hours because I'm not on the max 20 plan.

And maybe for yourself, you might actually either want to rechange what model you use here.

But anyway, I kind of want to make a demonstration now of how powerful this is.

So like I said, I've got an empty folder here with nothing in.

I actually want to make a simple app, but I'm going to tell it I want to use better auth, which means it will have to probably look at external resources to find what that is.

Dependency, so I'm going to give a very vague build step so it can find all the information and build it a correct and I want to kind of show you the power here.

All right, so I wanted to build a better auth app.

I want to use tan stack start for this one, and I want to make sure it can work on Cloudflare.

So we want to make sure it actually works on Hono if possible.

Can you build this simple application which will just be with a landing page, an auth page and then a dashboard?

All right, so I'm going to press enter now and you should see what really goes on here.

All right, so let's see what this can do.

All right, so we can see oh my open code is here.

All right, so it should start planning all of this.

It can ask us a few clarifying questions, so we'll see what comes through.

All right, I'm just clarifying a few things for them.

I'm just telling you to use tan stack start, better auth needs to run in Cloudflare workers, some email and password is fine with auth, auth can come later and I just want to use drizzle whatever database it wants to use.

So I'm just literally giving it the chance to make all these decisions for itself.

So let's see what happens.

All right, so you can see here it knows it's quite a different multi-step task.

It's got a lot of different things to do.

It actually needs to do some research.

All right, so it seems to know exactly what it's going to do.

I'm going to ask it.

All right, so I kind of forced it a little bit here, but this is probably would work with Opus on it, which I've seen in the past, but you can see here we've got this background task working.

We've got the librarian and it's telling us, hey, research, auth step and prompt, find better auth documentation, real world examples, things like that.

And we can also see here librarians also research tan stack start Cloudflare deployments so it knows how to deploy.

And also we also got another trust like research for the drizzle Cloudflare d1.

So it's got the three librarian agents running in parallel on the research, bit of auth and drizzle rrm.

So we'll see what comes back.

And it's going to wait for that.

We will only see pop-ups when it's done.

So literally we will get like a toe saying research is done.

If I go next, we can go see what's happening in the background.

So this is one of the librarians.

So we can see it's looking at the repo.

Yeah, it's trying to clone it in so we can grab this information.

Then we check so I can also go to the next one, what's happening here.

So this one is about the tan stack start Cloudflare.

So we're grabbing that.

And if we go to the next one, this is back to the main.

All right, this is we're back in the main one.

We're just waiting for the sub-agents to finish.

So we can see here we've got this background task is complete, the research.

And we've got another one that also came back.

So our research is slightly coming back.

So we can see if we had three tasks running in parallel doing the research, and it all came back quite nicely.

All right, so it's got the comprehensive stuff now.

It should be able to start building.

Are we still waiting for the last ones?

Pull it.

Okay, so all three tasks are back now.

It should now carry on and build from here.

I chose this one because it's kind of a stack that's not really well known.

So it should actually have to look for some documentation to really get going.

We can see it's just cloning some of the examples it got.

One thing to note, I don't know what happened here, but it kind of got rid of my .open code somewhere along the line.

my .open code somewhere along the line.

So I think it tried to clear the repo to make it a clean install, and it kind of just wiped it out a bit.

So I don't know what happened there.

Very, very interesting.

All right, we can see here it actually had the system reminder.

This is from a hook.

There to do confirmation status zero out of 10 and completing.

So it needs to quickly do it.

We can see here it defaulted back to Opus 4.5 regardless.

So the model here that you see is not what it's using.

It's actually still using my cloud Opus 4.5.

Come on.

All right, we'll see here it's actually compacting because I'm at 87% here.

It will then reduce its context and carry on going.

So this is actually an added feature in the back of how it works.

So we can see it reduced down to 77% and it's just going to carry on going.

All right, so we can see our context is reset here.

It's going to do some research as well to kind of figure this out.

Again, like I said, this is a very difficult task in the beginning.

This is not really well-known infrastructure and a stack app that is well-documented.

It's very niche.

It's probably very new.

So I actually am testing it to the limits here.

If I gave it something a lot more standard, it could probably build it a lot quicker.

But I want to kind of show you how good this is just by querying around.

So we can see here it's actually using its MCP, the GRIP app there, to search different GitHub repos to kind of get different information.

different information.

So very, very awesome to see how that works under the hood and where it stores it.

So just in case if you are wondering where it stores all this, this is all standing the temp folders behind the scenes.

So it'll clear as things go.

So you guys don't have to worry about it.

And this is all built in from the get-go.

So now it's going to try to use that information and fix the errors that we got.

All right, so we can see it's doing quite a few different things.

All right, so it's going to try to do these both, install and then fix.

All right, let's see what it's made.

This is open this cursor browser.

All right, so nothing too big, but we didn't give it any style.

This is literally all it has at the moment.

All right, I'm just telling it it needs to prove the front end because it doesn't look good.

All right, so it finished our UI updates.

So we will check that.

I don't think it's going to work off the back.

So this is C.

Yeah, OK.

So let's just run our server again.

All right, so we can see it's built a little bit of a better looking application now.

The dashboard has just got a bit of loading and everything else looks a bit better.

All right, so this is C.

If it what it actually built here.

So if you go to drizzle, this has got pretty simple things.

Nothing really here.

So it's tried to make a schema just of the user information.

And if we go to the file, we can go lib, we can go DB schema, we go with clients to see how it all works.

We've got auth.js running through there, drizzle adapter, you name it.

All right, so I'm actually going to ask it to help me just install the D1.

So I actually want to see if it can do this.

I've not done this on my open code, but essentially I'm just telling it to do that.

I'm just telling it like, hey, I've just logged into Cloudflare.

This create the D1.

All right, so this is asking me to run this.

Let's just do that quick.

So I'm just going to hit this clear.

I'm just going to run the command it's telling me.

This will just tell me to log into Cloudflare.

You are going to see this because I'm going to do this in my own account.

But essentially I'm just granting it access to my Cloudflare.

So you can then instantiate the resource.

All right, and then if you go through it, you'll see I've kind of done something like this.

You've given authorization to the Wrangler.

So great.

So it's there.

I'm going to tell it it is done.

All right, so now it should be able to create the D1.

All right, I'm going to ask it to test a login after it's deployed everything to Cloudflare.

So this is where our new app is.

So if I actually go here, so if I actually show this and if we actually go there now, so this should be where the app's deployed.

So this is our app deployed on Cloudflare.

We've got the dashboard.

Nothing really happens there.

We've got the sign in.

All right, so we're just updating the bit of auth to work a bit better.

You know, we can always change the secret and production to our own thing if you want to be secure, but this is just a demo purpose.

All right, I'm going to try this out.

So we're going to go back here.

We're going to say create a new account, which does not work.

Okay, so all right, so I'm just telling it to make sure the create account is there and there's a flow that we can use because it's not really there.

All right, it's doing compression again because of that 87%.

All right, we can see there's an issue with it trying to connect to the Cloudflare database from the Cloudflare workers.

So it's actually just looking at some docs to see if it can figure out what's going on.

Hopefully you guys are starting to see the power of this.

It does a lot of different things in the background to process to do a task quite quickly.

So this is very helpful to doing very complex, well-structured tasks that you know what you're doing, and you just need to keep eye on what is going on.

I wouldn't recommend using this plugin for simple tasks.

It seems to maybe sometimes go a bit nuts.

From my experience, maybe that will be changed in the future updates.

But you know, my experience with it does vary.

So like you can see in this one, it was very slow at the start.

Now it's starting to pick off a lot more at the end, which is interesting.

All right, so it is complete, which is great.

So we can actually test this out.

We're going to go back to our site share.

We're going to just run it.

So this is how it looks.

We've got the sign-in.

If we go to create a new account, so it built the API, but it did not build the front end.

All right, so I'm just going to give it some clues.

All right, so for the first time here, we can actually see it using the LSP diagnostics here, trying to find out what's going on.

There's no issues coming back.

All right, so it is fixed from this check again.

Get started.

Okay, it's coming to us.

It's creating a new account.

Let's just go test.

And we're just going to test again here.

Test.com.

And I'm just going to make that the same password.

I'm going to sign up.

Cool.

There we go.

Session data expires, everything like that.

And that is our sign-in.

We can sign out.

And this is the dashboard.

And home dashboard.

And we can sign out again.

Dashboard access denied.

Sign in.

Cool.

So we can see it is actually working quite nicely.

Again, this is all deployed on Cloudflare.

So this is actually a working website right now with authentication baked in and we're using better auth.

All right, so yeah, that's a demo of Oh My Open Code.

As you can see, it actually is quite powerful.

It's got all the little sub-agents and it uses its MCPs to its full disposal, especially using context seven and that grip app to get relevant information.

So you can actually progress with the task, even though it had very little information to go with from the get go.

As you can see in our example, we built a very simple app that we deployed on Cloudflare.

So it was actually working on the internet and it actually managed to grab the right document and sort of the issues that are running, even though it did not know how to actually build it properly.

So it was really good to see that it worked.

It was a bit slower, but hey, it was a complex task.

I didn't expect it to do as well as it did.

So very happy with that.

One thing I do want to note, if you are thinking about using it, it does use tokens at an alarming rate because it can run things in parallel.

Really try, you know, specify your configuration correctly.

configuration correctly.

I did see some odd behavior as you saw in this example.

It deleted my dot open code in the local project, even though it was never prompted.

That was a bit odd, but hey, if you're using git versioning, you can just always undo and then carry on.

But that's one thing to note.

It is a bit more complex, so some things can go a bit wrong.

But for the most part, it is a really good plugin.

It does a lot of things very well.

It handles complex tasks because of this researching tool that it can use and form things really well.

And we can see it did it quite well.

We also saw it actually instantiated the D1 and initialized beta auth for me from the get go.

Anyway, this was just a simple example I thought I could use to show you the power of it.

And if you enjoy videos more like these, please hit that subscribe button.

And thank you guys so much for watching.

And see you in the next video.