I Built NoteOcean So I Could Stop Searching Through My Notes

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

Ever remembered writing something down but couldn't remember which page it was on?

I built NoteOcean — an open-source, AI-powered note-taking workspace that lets you ask questions about your notes instead of manually searching through them.

NoteOcean combines a Notion-inspired editing experience with an AI assistant that retrieves relevant information from your saved pages and generates answers with source references.

Here's what it includes:

• Real-time Markdown editing: Type # and watch it transform into a heading as you write.
• Slash commands: Insert headings, lists, quotes, and editable code blocks using /.
• Instant page search: Find notes by title with keyboard-friendly navigation.
• Local-first persistence: Create and edit notes without login or registration.
• NoteOcean Bot: Ask natural-language questions and get answers grounded in your saved notes, with links to the relevant sources.

The idea is simple: your notes already contain the information you need. Finding it shouldn't be another task.

Demo

Live application: https://noteocean-l222m.ondigitalocean.app/

Code

GitHub repository:

Golixco / NoteOcean

NoteOcean 🌊

A modern, fast, AI-powered personal workspace inspired by Notion, featuring instant Markdown transformations, slash commands, real-time title search, browser-local persistence, and an AI assistant strictly grounded in your personal notes.

🌐 Live Application: https://noteocean-l222m.ondigitalocean.app/
📄 Hackathon Presentation: View Slide Deck (PDF)

---

📸 Screenshots & UI Preview

NoteOcean Workspace Canvas	Grounded AI Assistant (NoteOcean Bot)
Fast, zero-login workspace featuring instant Markdown transformations, editable code blocks, and real-time title search.	NoteOcean Bot answering queries strictly grounded in saved personal notes with clickable source citations.

---

📊 Presentation

The official NoteOcean hackathon slide deck details the problem statement, product features, editor interactions, grounded retrieval architecture, privacy boundaries, open-weight Google Gemma integration, and DigitalOcean App Platform deployment.

• 📄 View the NoteOcean Hackathon Presentation (PDF)

---

🌊 Overview & Product Vision

NoteOcean is built for software engineers, students, and writers who want the rapid, keyboard-first velocity of Markdown coupled with rich block layouts…

View on GitHub

The project is open source under the MIT License.

How I Built It

I built NoteOcean using Next.js, TypeScript, Tailwind CSS, and Tiptap/ProseMirror for the rich-text editor.

Pages are stored in the browser using localStorage, so users can start taking notes immediately without creating an account or setting up a database.

The AI workflow is the interesting part:

1. The user asks a question about their notes.
2. NoteOcean searches locally stored content and retrieves relevant snippets.
3. The question and selected context are sent to a server-side API route.
4. The Backboard API routes the request to the configured open-weight Google Gemma 3 27B Instruct model.
5. NoteOcean displays the answer alongside references to relevant pages.

The API key stays on the server, and the assistant is instructed to acknowledge when information cannot be found rather than inventing facts.

The AI integration requires an internet connection, and relevant note excerpts are sent to the configured provider when a user asks a question. Notes themselves are stored locally in the browser; they don't automatically sync across devices.

Why Does Open Innovation Matter?

I wanted the AI behind NoteOcean to be more than a black box hidden inside a proprietary application.

Open-weight models give developers more flexibility to experiment with model choices, adapt prompts, and build workflows around their specific needs.

For a tool like NoteOcean, that flexibility matters because the AI should work with the user's own context rather than provide generic answers.

Using an open-weight model also means the application isn't conceptually tied to one model forever. The underlying model can potentially be changed as requirements, costs, and capabilities evolve.

I'm using Backboard as the API layer for the current implementation rather than running inference locally, so this version isn't an offline AI application. My goal is to demonstrate how open-weight AI can make everyday software more useful while keeping the architecture small enough for an independent developer to build and maintain.

The Lazy Payoff

Imagine writing down an address, a deadline, or an idea a few days ago.

Normally, you might open several pages, search for keywords, scan paragraphs, and piece the information together yourself.

With NoteOcean, you can simply ask:

"Where did I write down that information?"

The assistant finds relevant notes, answers the question, and points you to the source.

No need to remember the exact page title. No need to manually inspect every note. Just ask and get back to what you were doing.

That's the kind of laziness I wanted to build for: not avoiding useful work, but removing the repetitive work that gets in the way.

Prize Categories

• Best Use of DigitalOcean — hosted on DigitalOcean App Platform.
• Best Use of Gemma — using a Gemma-family open-weight model for question answering.
• Best Use of Backboard — using Backboard's API for the AI workflow.

Closing Thoughts

NoteOcean started with a small idea: what if our notes could help us find their own answers?

Building it during Hacktoberfest has been an opportunity to turn that idea into a working application and explore how open-weight AI can improve a familiar everyday tool.

The goal is to make knowledge easier to use, one question at a time.

NoteOcean — Make your notes think with you.