- About
- Live Deployment
- The Problem
- Features
- Privacy First by Design
- How the AI Works
- Tech Stack
- How It Works
- Getting Started
- Usage
- Project Structure
- Roadmap
- Why Open Innovation Matters
- Contributing
- Acknowledgements
VaultLedger AI is a bookkeeping and invoice sorter built for freelancers, designers, developers, writers, and consultants who want the speed of AI without handing their clients' financial data to a third-party cloud.
Drop in invoices and receipts. An open-weight AI model reads them, extracts the key details, categorises each one, and builds a clean, organised ledger, all with your data staying on your own device.
The app is wrapped in a premium black-and-gold interface with an interactive 3D vault, because managing money should feel secure and look good.
Built for the Hacktoberfest Open-Source AI Challenge, Week 2: Side Hustle.
VaultLedger AI is deployed and ready to try right now. No installation needed.
Freelancers deal with sensitive financial information every day: client names, payment amounts, tax details, and bank references. The usual options all have drawbacks:
| Option | The catch |
|---|---|
| Manual spreadsheets | Slow, tedious, and easy to get wrong at tax time |
| Cloud AI bookkeeping tools | Your invoices are uploaded to someone else's servers |
| Doing nothing | Receipts pile up and tax season becomes a scramble |
For freelancers working under client NDAs, uploading invoices to a third-party service can also be a contractual risk. VaultLedger AI is built so that risk doesn't exist.
- Upload invoices and receipts and let the AI extract vendor, date, amount, and tax details
- Automatic categorisation into income, expenses, and tax-deductible items
- Client-wise organisation so you can see who you earned from and what you spent
- A structured ledger generated from your documents
- Running totals and summaries at a glance
- Easy to review, correct, and keep up to date
- Local-first architecture: your documents and ledger stay on your device
- Core sorting powered by an open-weight model you can inspect and control
- No required account and no required cloud upload
- A black-and-gold 3D vault interface that makes your financial overview feel secure and tangible
- Responsive, premium design built for both desktop and mobile
- Opt-in Gemini integration for users who want faster responses and accept the trade-off
- The core experience works without it
Privacy is the foundation of this project, not an afterthought.
- Local-first: your financial data is processed and stored on your device.
- Open and inspectable: the core AI is an open-weight model, so you can verify how your data is handled instead of trusting a promise.
- Opt-in cloud: any cloud-powered feature is optional and clearly separated from the core workflow.
- No lock-in: your data stays yours and can be exported.
VaultLedger AI is built around an open-weight model that runs under your control.
- Read: the model reads the text of an uploaded invoice or receipt.
- Extract: it pulls out the key fields: vendor, date, amount, tax.
- Classify: it assigns a category (income, expense, deductible, and so on) and a client.
- Record: the result is written to your local ledger.
Model used: [your open-weight model name], run via [your inference method, e.g. Ollama / llama.cpp / hosted open-weight endpoint]
Because the model is open-weight, you can swap it for a newer or better one without rebuilding the app.
| Layer | Technology |
|---|---|
| AI core | Open-weight model (gemini) |
| Optional AI layer | Google Gemini (opt-in) |
| Frontend | Web app with an interactive 3D vault, designed in Google Stitch and built in Google AI Studio |
| Backend | Invoice-processing and ledger logic, built with Claude |
| Storage | Local-first (on-device) |
| Deployment | Vercel |
| Research | Perplexity |
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Upload │ → │ Extract │ → │ Classify │ → │ Ledger │
│ invoice or │ │ vendor, date │ │ category and │ │ totals and │
│ receipt │ │ amount, tax │ │ client │ │ 3D vault view│
└──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘
all processed locally by an open-weight model
No setup needed. Open the live demo:
👉 https://hacktoberfestweak2.vercel.app/
Installation
# 1. Clone the repository
git clone https://github.com/[your-username]/[your-repo].git
# 2. Move into the project folder
cd [your-repo]
# 3. Install dependencies
npm install
# 4. Start the development server
npm run devEnvironment variables (optional)
Only needed if you want the optional Gemini layer. The core app works without it.
# .env
GEMINI_API_KEY=your_key_here
⚠️ Never commit your.envfile or API keys to the repository.
- Open the app and enter the vault.
- Upload one or more invoices or receipts.
- Review the extracted details and categories the AI suggests.
- Correct anything that looks off.
- Browse your ledger, totals, and client breakdown.
Update this section to match your actual folders.
[your-repo]/
├── src/ # Application source code
│ ├── components/ # UI and 3D vault components
│ ├── lib/ # Invoice parsing and ledger logic
│ └── styles/ # Black-and-gold theme
├── public/ # Static assets
├── .env.example # Example environment variables
└── README.md
- Invoice upload and AI-powered extraction
- Automatic categorisation and ledger
- Interactive 3D vault interface
- Live demo deployment
- Export to CSV and PDF
- Tax-season summary reports
- Multi-currency support
- Recurring invoice and subscription detection
- Fully offline desktop build
- Support for more open-weight models
Have an idea? Open an issue and let's talk.
A bookkeeping tool is only as trustworthy as its handling of your data.
With a closed API, every invoice goes through someone else's servers, under terms you can't audit and can't change. Open-weight models made this project's core promise possible: your financial data never has to leave your device.
- 🔍 Verifiable privacy: you can inspect and control the model.
- 💸 No per-call costs: local inference doesn't bill per invoice, which protects freelancer margins.
- 🔄 No lock-in: swap in a better open model whenever one appears.
- 🤝 Safe for client work: usable under NDAs without breaking agreements.
For a financial tool, that's the difference between "trust us" and "check for yourself."
Contributions are welcome, and this repo is a good fit for Hacktoberfest.
- Fork the repository
- Create a feature branch:
git checkout -b feature/your-feature - Commit your changes:
git commit -m "Add your feature" - Push to your branch:
git push origin feature/your-feature - Open a Pull Request
Good first contributions:
- Improving invoice field extraction accuracy
- Adding support for new invoice layouts
- UI and accessibility improvements
- Documentation and translations
- The open-source AI community, for the open-weight models that make local-first AI possible
- DEV and Major League Hacking, for running the Hacktoberfest Open-Source AI Challenge
- Google Stitch and Google AI Studio, for the frontend design and build
- Claude, for backend development
- Perplexity, for research
Built with 🖤 and ✨ for freelancers who value their privacy.
⭐ If you find this project useful, please consider giving it a star!
