Flow State 2.5 Initialization Tutorial 🚀

Hey there.

Abe Burnett, founder of Flow State.

And today I'm going to show you how initialization of Flow State works.

I'm going to be showing you how it works with the most recent release of the SE edition.

So version 2.5 is what we're going to initialize.

you can see that this is a really empty folder, so there's like nothing in here other than just Flow State.

and so it's just for purposes of trying to keep this very time constrained.

So first of all, I'm going to make sure the git is initialized so I can show you the Git functionality.

Okay.

And then Git's initialized or do Git status.

Okay.

Then I just come into this folder, I open up quick start, and then I get the prompt here.

And then I like to use Claude for this because I find that Claude works just generally really well.

I mean, it was really good under Sonnet 4, point 5, and it's gotten even better in general with Opus 4.5, which is now the default in Claude code.

so highly recommend Claude if you can afford it.

I actually have ChatGPT Pro and Claude, or what do they call it?

Yeah, Claude Max, the a hundred dollar plan per month.

And then I have GitHub copilot.

I think that's $10 a month.

it is cheap.

It doesn't work as well as Claude or Codex.

so just keep that in mind.

Even though it's the same models, it's the weirdest thing, the same models, working with them through GitHub Copilot, not remotely the same experience.

So just keep that in mind.

All right, so I'm going to close this terminal to give us a little more room to work with here.

Like, my window gets very full of all these AI windows, so we're just going to drop that prompt in there and say, go.

Excuse me.

And so one thing to keep in mind is that this will be relatively fast because this is an empty directory.

So there's not a lot for flow state 2 and the AI to have to kind of peruse and browse and think about.

The bigger, the full the project that you drop flowcade into, the longer it will take to get set up because there's just more for it to review, more context for it to gather and kind of triangulate from to get a sense of, the environment that it's going to be operating within.

Just so you know.

So, I mean, if you're, you know, if you drop this into a big project and it takes a little while to initialize.

That's probably to be expected.

Just as this works.

I will say that other things I really like about Claude, is it seems to be, to me, I really like its balance of verbosity and how it tells me what it's doing.

It also has really nice ability to interrupt it.

So if you see it kind of going off the rails, doing something where you're like, that doesn't seem quite right, you can jump in and, and be like, and feed it more stuff.

And it circles back and kind of checks to see if there's anything it needs to kind of fold into what it's thinking about.

All right, so what do we got here?

Lot code, new project GitHub, detected.

So we're going to say I think a lot of people will like this kind of private function.

So that means that it does not commit anything to git, it's completely hidden as far as the repo is concerned.

and it uses kind of a little known functionality to do that by putting things under Git info exclude.

And anyway, and then it also obfuscates the file names to make it even more private.

So we're going to show you that.

We're going to say yes to obfuscated file names.

And so one of the key changes, in this edition of Flow State is that it actually puts all the files under a Flow State directory, which helps avoid conflicts with say like a Git readme MD file.

And so we're going to say yes.

So, just so you know that that will be kind of a folder.

It will be hidden in this case.

but otherwise it just kind of helps keep your project folder a little bit more tidy.

Okay, so we're just going to go through here question one.

The other beautiful thing about Flow State being kind of like a layer on top of AI is that you can reason with it.

So at any point during this process you can ask questions, you can give it more context.

This is not a rigid kind of UI in the sense that if you have an app on your phone, if the app doesn't work the way you want it to, you're just kind of stuck with AI interfaces like this for AI.

They're malleable.

So just keep that in mind.

So right now I'm going to say a, get Flow State completely.

I mean, you know, I'm gonna say that because I don't have a, like a real project in here.

I'm just say D for now.

But again, just Keep in mind that a lot of times when I'm filling this out on an actual project, I put a lot of context into these questions, as much or as little as I have at the time.

Because sometimes you don't know what your deployment target is going to be, sometimes you don't know kind of the subscription tiers you're going to have.

But it can be useful just to give in flow state any amount of context that you have.

Right.

initialize.

Cool.

Now it's going to walk us through some settings.

So scaffolding level.

So this is to teach us kind of how to do five coding, you know, AI orchestration.

I'm going to say one for maximum teaching.

You can always change it later by saying something like change scaffolding to whatever.

You could even ask it.

If you forget the phrase, you can say like hey, how do I change the scaffolding level?

And then it will give you, it will just kind of help you out.

Right.

quality mode.

I'm going to say methodical again.

You can change it just by saying change quality mode to or a variation thereof.

So it's going to put it all under C doc.

So it's going to put adapters under the main directory because that's where they need to go.

But otherwise it's going to put everything under flow state.

But you'll see these new adapter folders.

I don't really talk a lot about cursor support.

I know a lot of people like cursor.

I haven't really used it, but I did work with AI to try and anticipate what cursor would need, to make flow state work for it.

So I have that there in case someone wants to try, using flow state with cursor.

But basically flow state just takes like a, an approach of kind of assuming, well, you know, we try to be as prepped as we can for almost every platform and then whatever you want to be using, that's up to you.

But we want to be as flexible as we can.

Honestly, any kind of platform that has kind of an agentic kind of vibe coding interface where you have kind of these, either cli, like terminal kind of windows or a window like this where you're interacting with an agent in an agentic context, it should work fine.

you might have to like when you do initialization, you might have to kind of bump it, nudge it and be like, hey, you know I'm using you and in Juno or something.

I think that's what it's called or the Jetbrains product.

Project.

Product.

Whatever it's called.

but you could say like, you know, I'm using you in Juno.

Please adapt all the flow state kind of orchestration to work for Jetbrains Juno.

and it will comply again because it's flexible.

Okay.

We're not going to quite make the 10 minute mark, but we'll be close.

Also, while this is working, I will point out a few files you will care about.

if and when you get the chance.

You might enjoy kind of reading through more of the Quick Start file to understand.

Kind of like there's kind of the most important stuff is in here.

So for example, if you have any issues at the very top, there's kind of like a what to do emergency Flow state not responding.

then you can say just drop in this command, right?

This prompt read troubleshooting md Follow its AI instructions.

if you can't access this file, search this workspace and report what you find.

So it's basically intended to help you get going.

This is not so much an issue with Claude code or codex or even Gemini 3 Pro.

Gemini Pro.

I don't know how they say that.

if you're using like atmosphere from Google, it is more of an issue in like GitHub Copilot where for whatever reason the same models in GitHub Copilot will seem a little bit bone.

Like a little bit boneheaded sometimes like just a little dense.

So just keep that in mind.

But yeah, so it kind of walks through the high level stuff and then the other files you probably might care about is troubleshooting.

Some help stuff in there.

Reference.

This is kind of intended, it's called reference because it's intended to be used as a reference and not by just by you, but also by AI.

If you're upgrading from a previous edition, you can look at this file.

yeah, this is all the stuff it creates that it actually uses.

It's all in the Flow state.

Flow state folder.

All right, so we've got version 2.5 of the SE edition initialized and it's kind of given us some priorities, some configuration, verification complete.

It does a double check to make sure it did everything.

and then we can do.

Ta da.

Like it's hidden, like it's So you don't see any of this in git.

All right.

And so then for example you could just open a new window and say flow state or I prefer resume flow state.

But you know, any variation thereof will work.

So this is kind of like useful if you have to open a new thread, right?

And it will pull all of your context from all the plans, all that, the session notes that, the tasks notes, and all of these files here.

That's how it kind of handles and improves, and supports these models in their compaction cycles.

So like Claude code has a compaction cycle, so does GitHub, Copilot, so does Codex.

I, think also Gemini 3 Pro does as well.

But the compaction cycle can be a little iffy.

And so, basically with the models doing their best to compact well and then coupling that with what flow state does, then you get like the combination is really effective.

And it's actually something where I've seen that flow state is.

No matter how good the models get, flow state just works better.

So it's, it's really like, it's really interesting that I don't really feel like their models will ever really get good enough to not need flow state.

The only thing is that maybe some of these providers might come out with their own version of flow state, in which case that might be, an issue for me.

But we haven't seen that happen yet.

And I think that's partially because they are trying to solve a very broad set of problems for a variety of people.

And you know, they're not just focused on developers, or specifically like the, the folks that I'm focused on, you know, they're focused on a very broad audience.

All right, so it basically it's resumed.

It shows the current state, next tasks, that kind of thing.

So that's it?

Well, we didn't quite hit 10 minutes, but 16 minutes is not bad.

Thanks for watching.