Show HN: I built "Docker for code", isolate AI logic into semantic containers
I've been pretty annoyed by the "all or nothing" approach of most AI coding tools, which are useful, yes, but also risky. If i tell an AI to "fix this bug" it may probably do it, but it can (or will) hallucinate a fake library, mess up my manually optimized code or subtly delete important code without explaination or notice.
Glupe summarized is this: "Stop giving AI root access to your code, Glupe isolates AI logic into semantic containers, so your code stays safe".
Instead of sending your whole file to the LLM and pray it does not break, with Glupe you can surgically tell the AI were to write code and what to do using the $${ logic }$$ syntax. Code outside the container is preserved as is. Instructions inside the container are turned into source code and embedded into the output file.
Containers can be named and are cached and hashed, Glupe detects which containers changed in a source file; when you run `glupe project.txt -make` it only calls the AI to update the "dirty" logic containers and uses cached code for the clean ones, saving time and API calls and allowing for incremental builds using AI.
What if there is a syntax error on the output code? Glupe has a self-healing loop which feeds the error history back to the LLM and tells it to fix it. No more trivial debugging.
Why do I say it is "Docker for code"? It is a metaphor, it applies the same core principle of Isolation to source text that Docker applies to running processes.
In Servers: If you run an app directly on a host OS, it has "root access." If it behaves badly, it can delete system files.
In Coding: If you give an AI access to your file, it has "root access." It can "crash" your architecture by deleting manual code or changing signatures. Docker protects the OS from the App. Glupe protects your Architecture from the AI.
Docker solved "it works on my machine", Glupe aims to solve "
I wrote a white papaer on the subject which is available in the repo.
Any feedback or critcism is welcome and encouraged.
repo: https://github.com/alonsovm44/glupe white paper:https://github.com/alonsovm44/glupe/blob/master/WHITE_PAPER....