To improve my own practices and have more time to spend on my extensive studies, considering my preference of local first whenever possible, I obviously started by making my own coding VS2026 extension. Bare bones, raw, usable already. So, I decided to make an extra context assistant, introducing CodeFly. Yeah, I’m not the best at naming things…

I’m unlikely to physically share this project with others, at the moment at least. Today’s post is more to help others understand they can build cool things to improve their own work and productivity with personal assistants (LLMs). It has taken me a while to decide what to implement, and use, I took time to plan properly, then decided to start to finalise my prototypes and show off a usable solution I found.

CodeFly aims to document my projects as a wiki for myself, as well as my local code agents. So far, I’m still leveraging copilot for Claude Sonnet/Opus (usually), but I decided I should actively also take data privacy and safety into consideration. My personal hardware isn’t the top tier for massive LLMs, but at least I’ve started, and had decent success. I’ll eventually swap even more to local LLM use, just need to save up for the hardware.

So far, as a raw SQLite wiki using markdown without images, my LLMs can first document my code and practices for me. Then I can get my LLMs to inspect the projects and suggest improvements I can implement. Especially for data safety, user friendliness, and privacy focused solutions.

The first proof-of-concept wiki task was just to start the section for AeF; then to analyse the documented project and offer improvement suggestions.

Using my own branching, and personal code reviews, always seems to miss a ton of things I should have considered. Also, yes, it instantly reminded me it’s lazy to use static DB references.

When I ask copilot to follow context from CodeFly it does take a bit more time; but I’m getting broader fixes and movements sorted way quicker. I should probably make a user-friendly interface on the website view of the wiki for changing the pages like I did for my Above Average Personal Wiki, but I have to share I’m rather just going to add CodeFly as an extension to my wiki.

I can document all my code projects extensively now. Then soon use my CodeFly to move code between projects. Add features I’ve prototyped in the past (which died after prototype). Then, lastly release the projects that have constantly gotten far too much feature, and scope, creep. That’s the reason I get near a release, then hold it off for a few months, I start to “improve” whatever I notice in need of improvements.

First, the 3 critical issues found by CodeFly review, were fixed in around 15 minutes. I’ve seen news sharing how being a software engineer is moving towards using models for faster coding, with extensive review processes, and it feels alright to head in that direction. This felt more productive than manually reading through every code file, yet again, which might be missed otherwise.

I do feel it’s more important to still also spend as much time writing our own new code, to engage our minds and learn new practices or upgrades in languages, as a whole. It’s also important to learn what your LLM use can teach you, as well as make faster large-scale changes in your code base, when needed.

The above feature additions on the Wiki took only a moment, or two, while writing this blog post. That already feels like a productivity gain, only small additions I used CodeFly to rapidly implement to improve itself, took seconds instead of manual search and implement.

I feel this was a great idea to start to speed up my development process. I’m definitely enjoying learning way more new code instead of having to spend extra time reading through hundreds of files to remember where to implement things.