AquaAgent

Inspiration

The effect datacenters have on an environment's water supply and a town's electricity rates is worrying because it affects the ecosystem as a whole. We wanted to help mitigate this effect on our communities, so we created AquaAgent.

What it does

AquaAgent reduces token usage by caching similar prompts to save tokens (this works for all users on a repo) and making the user aware of how much water and electricity they save and editing their messages to mitigate token usage. Studies show positive feedback often leads to better results.

How we built it

For the backend, we employed Vercel for a serverless backend, LiteLLM for handling the text between the user and Claude Code, an OpenAI model for semantically embedding the text, and Upstash Vector and Redis for searching a created vector library and making sure the same question wasn't asked twice. For the frontend, we used Claude Design and Canva.

Challenges we ran into

Honestly, combining the backend and the frontend was incredibly difficult, because with such an elaborate backend, it was hard fitting it all onto one frontend webpage. We ran out of tokens multiple times too.

Accomplishments that we're proud of

We made a tool that actually has an impact on the world right now. Anyone can use it, and it doesn't cost anything. It's totally free to be eco-friendly, and making something that shows that is something we're proud of.

What we learned

We learned how important it is to act against water usage in data centers and how important it is that we do so as a community. Also, we became familiar with using Upstash and LiteLLM.

What's next for AquaAgent

We want to focus on the UI, making the user experience as positive as possible. Also, we hope to add account-specific details, focusing on adding a sign-in feature and the ability to track specific sessions within GitHub repos. Lastly, we want to implement this within places of education and employment, helping them optimize their usage and focus on sustainability without hurting their workload.

Built With

• canva
• claude
• litellm
• next.js
• react
• upstash