CAPYCODE

Inspiration

With the rise of artificially generated code and vibe-coding tools, we find that developing and debugging projects especially as a beginner is a challenging task. The lack of context of generated code may be too niche or complicated for beginners to understand which can even make some of the simplest projects seem overwhelming. That's why we created CAPYCODE.

What it does

CAPYCODE is a tool that allows you to verbally communicate with an AI agent as you develop code. CAPYCODE will take you through coming up with a project idea, reasoning through which framework is best for you, and coming up with a detailed plan and outline for your project. It will then provide you with a file editor, file explorer, a realtime output renderer, and provide you with context and support all throughout your coding process.

How we built it

Our stack:

• ElevenLabs
• Agentuity
• Google Gemini API
• Sandpack
• Convex
• Vite with React with Typescript

For our coaching feature, we leverage both Agentuity and ElevenLabs to create an multi-agent flow to make the user experience seamlessly.

Challenges we ran into

• Figuring out the best prompts and models to facilitate the agent to follow the pipeline
• Parsing the web structure and data to provide the agent with the proper context
• Learning Agentuity and Elevenlabs API. Since the agent mode is alpha it's have a bit issues and we have to implement some workaround with it.

Accomplishments that we're proud of

• The file editor
• Continuous conversation with the agent
• Context parsing and context memory

What we learned

• Create a new agent in Agentuity
• Use agent feature in Elevenslabs API
• Implements Convex with Auth0
• Use animation in pre-landing and landing-page

What's next for CAPYCODE

• VSCode extension functionality
• Support for Python and other languages
• Terminal support

Built With

• agentuity
• auth0
• convex
• elevenlabs
• gemini
• react
• typescript