Muser

Inspiration

Loveable inspired us to create a vibe coding agent that generates music and gave us ideas for the UX.

What it does

Muser lets users quickly generate music as code.

How we built it

We created the app using Tavily, allowing users to search for samples from various sites like YouTube. We used BAML to improve prompts and Gemini as the main model to generate code. Tiny agent MCP Servers were used to generate different beats asynchronously.

Challenges we ran into

Tavily web scraping was a big challenge—rate limiting made testing difficult. It was also tough to get a voice agent working and to configure BAML code. Most agent outputs were inconsistent, and sometimes the music generated was good, but not always.

Accomplishments that we're proud of

We’re proud that we were able to successfully generate music via vibe coding that actually sounded good.

What we learned

We learned how to develop a working, beautiful application that lets users create music on the fly. We learned a lot about Tavily web scraping and gained plenty of experience with music-to-code generation.

What's next for Muser

Muser should generate more consistent output, have a better agent workflow, and offer deeper integration with Tavily for samples users can add to their music.