Hey everyone!
I’M MAINLY TALKING TO THOSE OF YOU WHO ARE DEEPLY INTO VIBE CODING AND LOVE NODE-BASED THINKING AS MUCH AS I DO.
Together with a friend, we’ve built a tool called Flowcrest, and honestly it completely changed the way I work. I might be biased, but it solved a very real problem for me.
I’m the kind of person who loves planning ahead. When I start a project, I already have a full map in my head — ideas, structure, relationships, constraints, little details. And until now, I only had two ways to handle that:
1. Dump everything into one giant prompt.
2. Or hold myself back and develop in tiny pieces while taking messy notes.
Both are painful.
If I write everything into one big prompt, the AI struggles — not just because it’s long, but because the logic has no structure. Human language is terrible at describing a system with hierarchy, references, and flow. It becomes tangled fast.
If I break it into tiny prompts, most of my mental energy goes into keeping the whole structure in my head, planning ahead, and trying not to confuse the model. It’s slow and mentally exhausting.
And since I come from a background of using node-based tools (like Blender’s node graphs), one day it hit me:
Why not build my system the same way?
Visually. Modularly. With tiny parts to create a structure.
So we built Flowcrest.
How it works (this is where people usually ask questions, so let’s make it clear upfront):
Each node represents a small module of your project — a tiny, focused idea, written in plain human language. No code needed.
You give each node:
a simple description
optional input/output ports representing what information flows in and out
connections to other nodes to express the logic
You end up with a clean, visual “blueprint” of your project.
Then Flowcrest exports two things, which the AI uses together:
1. A general instruction prompt
This explains to the AI how to read your Flowcrest structure.
It tells the model what a node is, what inputs/outputs mean, and how the whole blueprint should be interpreted.
It’s the decoder key — NOT about your app’s content.
2. A JSON or TOON file describing your actual project structure
This contains the nodes, connections, descriptions — the entire logic map.
This is what the AI uses to generate the actual code or output.
Because it’s structured, it requires far fewer tokens, and the model understands it far more reliably.
This combination gives you:
clarity
modularity
better outputs
and no more losing track of ideas
Fast forward to today — and we’ve made it even better.
We launched new browser + IDE extensions:
You can now send your entire Flowcrest project directly to your AI agent with one click.
The extension automatically:
exports the structure
generates the correct decoder prompt
copies both into your clipboard
Just paste → send → continue building.
It also supports “diff sending”:
If you tweak something, you don’t need to resend the whole project.
The AI receives:
the previous version
the new version
and the semantic diff
So it instantly sees what changed and keeps the context.
This feature alone has been a game-changer.
It’s been only one month since launch, and surprisingly many people are already using it. You can try the free demo, and if you like it, a monthly subscription costs less than a cheeseburger menu (or you can support us with a lifetime access).
Based on user feedback we’re adding new things constantly — you can already draw on the canvas using a tablet, export only a sub-section of your project, and a lot more.
Next big feature: local project saving, so your ideas never leave your machine. (Also user-requested!)
Thanks for reading — and have a great day!