VersionContext

Our landing page!
A sample repo with branching and merging
You can modify your files, and/or prompt gemini and get direct modifications

Inspiration

AI is stupid... it makes mistakes in code, goes into infinite loops of thinking, and breaks things unexpectedly. Worse, if you're collaborating with other people on the same codebase, AI can make the same mistakes over and over again. As vibecoding becomes more and more mainstream, this is becoming a bigger issue. Our project aims to solve this issue by adding AI agent context to shared codebases with every push!

What it does

Our web app has GitHub-inspired functionalities, including features that allow users to create new repos, share repos, create new files and folders, and includes a built-in IDE that allows users to edit their code and create new files directly in the browser. When the user writes and commits code, with or without AI usage, our built-in Google Gemini API summarizes the context of the code, including any mistakes made, the context of when this commit was made, and more. This information is then conveniently stored in a graph/timeline of the branches and commits of the repo, readable by human users, and usable by the AI agents of other users working on the same codebase.

Our target audience is coders who utilize AI Agents/LLMs to write their code and collaborate in shared codebases.

How we built it

Frontend:

Next.js + React (TypeScript)
Tailwind CSS
Google Fonts via next/font

Backend:

FastAPI + Uvicorn (Python)
Pydantic v2 (data validation)
aiosqlite (async SQLite database)
GitPython (git operations)
Google Gemini API (google-genai) 0 for the AI/chat features

Infrastructure:

Docker + docker-compose
DigitalOcean
Cloudfare
SQLite (database via aiosqlite)

Tools:

Claude Code

Challenges we ran into

There have definitely been technical challenges - for instance, we encountered unexpected bugs and edge cases with branch merging when creating our GitHub-inspired functionalities. It took us a while to all be on the same page and understand what was going on with the branch merging, which branch was local and which was committed to our system, etc. In the end, we solved it through collective iteration and teamwork.

Accomplishments that we're proud of

We are very proud of our idea and our execution! We believe it's a very solid idea, and a quite useful one for coders as well - and we also believe our execution lives up to it. Our teamwork was very seamless, and we each contributed important components to the project.

What we learned

We learned a lot about how GitHub works (since our project has functionalities largely inspired by Git) - branching, rebasing, all that jazz, how to do context compression for AI models, prompt engineering best practices, and so much more. We also each deepened our expertise in our respective areas of knowledge - whether it is full-stack development, graph theory, backend, front-end design, or something else entirely.

What's next for VersionContext

We look forward to developing a desktop Electron app/integration with existing IDEs such as VSCode! Our product should be able to benefit coders around the world who frequently use AI, and we would be excited to make this happen.