rustsn – Open Source project for generating code and interacting with existing code through LLM
I've been working on creating a tool called rustsnwhich allows you to generate, compile and test code using LLM (Large Language Models). The original idea was to automate the process of writing small pieces of code – called snippets – for different programming languages based on explanations provided by the user. This tool began to develop and acquire new capabilities, such as generating complete code for applications and explaining existing code based on vector representations (embeddings).
When I first started working on rustsnthe main goal was to ensure that the user could, for example, simply describe in words what function he needed to write, and the system would automatically generate working code. I started with Rust because the language has powerful typing and testing capabilities, making it ideal for writing safe and performant code. Later I added support for other languages such as JavaScript, Python, and TypeScript.
This tool has become much more than just a code generator. For example, I added the command askwhich allows you to get an explanation of your project's existing code. To do this, a special model is used that analyzes the sources and finds the most relevant files based on your question.
In the near future I plan to significantly expand the functionality rustsn. The first thing I want to do is integrate Dockerso that users can run the tool not only on the host machine, but also in containers. This will simplify the deployment and use of the tool on various platforms, ensuring the stability of the environment.
In addition, one of the key goals is to expand the ability to generate code. Now rustsn does a great job of generating individual functions, but I would like it to be able to create full-fledged ones seeding projects – from command utilities to web services with REST API. This will make the tool much more useful for developers who want to quickly get the skeleton of an application based on a simple description.
And finally, to get the job done with rustsn even more convenient, I plan to add GUI as a web application. This will allow users to interact with the tool through a browser without having to go to the terminal, which is especially convenient for those who are not used to working with the command line.
I see great potential in the development of this project, and further steps are aimed at making rustsn a universal tool for generating and analyzing code that can help developers at all stages of their work.
The main challenge that I see in the development of rustsn is that the use of commercial, “front-line” AI, which requires constant cash costs for APIs and resources, will lose the fight with open-source solutions. We're already seeing a trend towards open models that can be run on your own hardware becoming more powerful and developer-friendly. That's why I chose to integrate rustsn with Ollama – it's a free and local solution that allows you to run LLM on your own computer, providing maximum flexibility and independence.
I am sure that in 1-2 years the optimization of LLM will make it possible to run them not on large GPU clusters, as is required today, but on ordinary laptops with relatively modest hardware. This will be a real revolution for developers: everyone will be able to run powerful AI models locally, without depending on third-party services and their tariffs. My goal is to prepare rustsn for this new reality by making it more efficient, autonomous, and accessible to every developer, regardless of their budget or infrastructure.
Thus, rustsn will not only make the process of writing and analyzing code easier, but will also become part of a future where every programmer will be able to use local AI solutions to create applications, analyze existing projects and automate routines.
If you want to take part in the LLM project and at the same time master the Rust programming language, welcome to my project: https://github.com/evgenyigumnov/rustsn
