Inkforge AI: 5 Minute Full Feature Demo ✍️

Hey there.

Abe Burnett, founder of Flow State AI.

And today I want to do a short video demoing, kind of the build for a full feature, I guess you could call it.

So this is a bit ambitious for a video and we'll see how long it takes.

But essentially for the app that I am building, which is a mobile first AI writing vibrating software called, inkforge, we are going to work on a context aware input bar.

So right now it just says, as you, as you can see in here, in here, forge the next scene.

Well, it's kind of generic and it would be more engrossing and engaging to, ask where the prompts kind of are related to what you see on the screen.

So that wherever you're going to have the AI generate text, the prompt and the bar relates to that.

So if it's kind of like the last thing is where one character of, say a female character is arguing with a male character, maybe the prompt is like, what does she say next?

so, so we're going to try and make that more contextually aware.

So, for example, what would it look like where you are mid action scene, current, forward to the next scene, but maybe then.

Or forward to the next scene instead of that.

What does Sarah say back?

so other key components are a narrative cursor tracks where you are, a scene cache where we have lightweight metadata per scene, a tiered prompt pipeline, UI behaviors.

And so it's broken out into a variety of implementation tasks.

And I think we're going to say, yes, we're going to go ahead and do it.

Please proceed with epic 18.

And so, just so you know, I'm going to be pausing this video and then restarting it kind of to cut out all the slow stuff, right where the AI is just thinking and compacting and thinking and compacting.

So we'll come back to talk about the meaningful stuff that it does and what Flow State does specifically.

But I think you're probably already seeing some things that Flow State plus AI Claude in this case does really nicely that I really like that if I have to work without, it drives me absolutely crazy.

So Flow State plus Claude is a beautiful combination.

I also use Flow State with Copilot.

the models there though, same models you might use with Claude code or with Codex.

but they don't work as well in Copilot as they do kind of natively.

but you know, I kind of work with all these different tools just kind of depending on what, you know, what's working for me at the moment.

Okay, so I'm going to pause this.

I'll be back when there's something more interesting to talk about instead of compaction.

but one thing you will notice is that with flow state compaction, not a big deal.

So between the models improving at how they do compaction and just the models improving period plus flow state, you know, you actually don't have to fear compaction because you can cross that gap without losing context and without things falling apart.

Like, you see that it, it did compaction efficiently and now it knows what we're doing and it just picks back up from the task list and the plans that are associated, that rich context.

Like it's just able to keep going.

All right, but I'm gonna, I'm pause this now and I'll be back.

Okay.

And now, you can see that we have finished T180.

It is checked to make sure that the files are all under the 350, lines of code cap.

It's just kind of a guideline.

Not, not a strict or rigid, and it's moving on to the next items on the to do list.

Okay, we're back for a second.

It is finished T182 and it's working on T183.

It's wiring the Mercury integration.

So I noticed that, one thing that was interesting here, it does this frequently, but it noticed that the file had exceeded the 350 lines of code cap.

And so it looked for an opportunity to proactively refactor.

And so it actually split out the mercury prompt, building into like a separate, importable, I guess that's what you'd call it.

Yeah.

So, yeah.

But I'll be back in a bit.

Okay.

And we're back and Epic 18 is complete where we have all four tasks completed.

Here's what was built.

So, the narrative cursor, scene cache entry, narrative cursor service, UI state machine, tiered prompt pipeline, and the Mercury integration with telemetry.

We use, Mercury because it's very fast to do certain things.

and speed is of the essence.

So it checks the different file sizes against our standards and it lets me know that the infrastructure for context aware prompts is ready.

It just needs to be wired into the writing mode screen for full integration.

So that's what I'm going to do next.

But there you go.

That's one feature in the bag.

ready to rock and roll.

So that's kind of what it is like to work with flow state.

Just.

Yeah.

So if you have any questions, check us out at the Discord Group.

I'm in there.

There's other members in there, happy to help and answer questions.

And also, you can also email me directly.

Abelowstateai.dev all right, talk to you later.

Thanks for watching.