Caret

Inspiration

I got inspired by Cursor and how it makes editing with AI super smooth. While I was editing my own research paper, I realized I needed something similar, something that helps with writing, editing, and fact-checking without constantly switching tools. So I decided to build my own version.

What it does

Caret brings AI into your document editor. It helps you write faster and better by:

• Showing AI completions as you type (like Copilot, but for writing)
• Letting you hit Tab to complete suggestions
• Checking if selected text is actually true using a built-in fact checker

How we built it

Tech stack used:

• Next.js for the app
• Lexical for the editor
• Tailwind CSS for styling
• Perplexity Sonar API for fact-checking and completions
• Drizzle ORM with PostgreSQL for saving docs
• React Query to manage saving and syncing

Challenges we ran into

• Making AI suggestions not mess up the text or cursor
• Streaming completions smoothly so it doesn’t lag
• Figuring out how to fact check only selected parts of text and show useful results

Accomplishments that we're proud of

• Got AI completions and Tab completions working in a way that feels natural
• Built a working fact checker inside a text editor
• Made it all work together in a clean, fast UI

What we learned

• How to customize Lexical for advanced features
• How to work with streaming AI responses
• How tricky it is to keep the UX smooth when mixing user typing and AI input

What's next for Caret

• Adding /commands like /summarize or /rewrite
• More formatting tools and themes
• Better version control and document history
• Eventually, live collaboration and maybe even a plugin store

Built With

• drizzle
• nextjs
• openai
• perplexity
• postgresql
• shadcn
• tailwindcss