Coco

Inspiration

I used to use this service called Augment code, which was an AI assistant built-in to VSC (my IDE) and would use my code as reference for offering assistance, reading through all my relevant files as context before helping me out and debugging. However, when they started charging and gave very small token limits for free rates, I wanted to make my own.

What it does

Effectively, it's a VSC extension that provides a chat area where you can talk to an AI model (configured to be Gemini 2.5-flash-lite right now) about your code. Future developments include integrating RAG to detect the codebase of the user.

How we built it

Microsoft has API documentation about VSC extensions, how to make one and what it can do. It was built entirely writing Typescript and making API calls to Gemini.

Challenges we ran into

Figuring out how to use both the VSC API and Gemini API took a lot of work, trying to sift through legacy and current versions of API functions.

Accomplishments that we're proud of

I was able to get a working VSC extension with a chatbox you can enter prompts into, and have an AI return a message about the prompt. It may not be what I fully wanted, but it is a very promising start.

What we learned

I learned how to integrate LLMs into my software, and plan to incorporate it more into my future projects whenever it could be helpful.

What's next for Coco

Integrating chat history so the LLM can have full conversations, implementing RAG so the AI can parse the codebase, fine-tuning the AI so that it is specialized for coding assistance and is aware of relevant coding files so it won't unnecessarily bloat it's context

Built With

• gemini
• typescript
• vsc