If you use an AI to help you code, you know the exact moment it fails.
It’s usually around 2:00 PM. You’ve spent the whole morning working through a logic problem. You and the AI have finally figured out how a specific part of your app should work. You feel unstoppable. Then, you open a new file to connect that logic to your database, and you ask the AI a simple question.
The AI looks at you (digitally) and says: "I don't see any database logic. Can you paste the code you're talking about?"
Just like that, the momentum is gone. You’re back to copy-pasting code blocks from one tab to another, trying to "remind" a genius model what you literally just did ten minutes ago.
I call this the Goldfish Effect. And I got so tired of it that I decided to build a fix. I call it Dalexor MI.
The Problem: AI has no "Big Picture"
We keep hearing about how smart these AI models are, and they are. But they have a massive flaw: they only see what is right in front of them.
Most tools use a simple search to find snippets of your code. It’s like trying to understand a 500-page book by only looking at the specific page you’re on and three random paragraphs from the middle. You see the words, but you lose the story. You lose the why.
When you're building a real project, the why is everything. You need the AI to know why you chose one library over another three weeks ago, or how a change in your "user" file is going to break your "payment" file on the other side of the folder.
How I built Dalexor to solve this (The simple version)
I didn't want to build another "chat with your files" wrapper. I wanted something that felt like a permanent brain for my project. Here is how it actually works:
- It watches in the background I built a small tool you run in your terminal (
dx watch
). It just sits there silently. It doesn't nag you. But every time you save a file, it takes a look. It’s like a silent partner taking notes on the project while you work.
It filters the signal from the noise If I record every single time I fix a typo or add a space, the AI's memory gets cluttered with junk. So, I built an engine that measures how "meaningful" a change is. If you just changed a comment, Dalexor ignores it. If you added a new login flow or changed how your data is structured, it says, "Okay, this is important," and it saves a "snapshot" of that logic.
It builds a map, not just a list Files in a project aren't just a list of names; they are connected. My tool maps these connections. It understands that File A depends on File B. When the AI looks at your history, it isn't just seeing a bunch of text files; it’s seeing a map of your entire system architecture. It knows how the pieces fit together.
Privacy that's actually private This was the most important part for me. My code is my life's work. I don't want to send my entire history to a third-party cloud where someone else can see it. So, I made sure that your code is "locked" on your own computer. It uses a key that only you have. By the time it hits the internet to be saved in your history, it’s already unreadable to everyone—including me. Your AI assistant can see it (because you give it the key), but no one else can.
Why this changes how you build
The goal isn't just to write code faster. The goal is to stop re-explaining yourself.
Since I've been using this, my conversations with AI have changed. Instead of saying "Here is my code, please fix this bug," I can say "Look at the changes I made this morning—did I break the way we handle user sessions?"
And because the AI has that history, it can actually give me a real answer. It remembers the decisions we made. It has context.
I’m finally at a point where I’m ready to share the documentation and the first build with other developers. It’s simple, it’s private, and it actually remembers.
If you’ve been hitting that same "context wall" and you're tired of re-pasting your work every twenty minutes, I’d love for you to check it out.
Top comments (0)