Local AI Document Scanner

Budgetzilla features a powerful local-first AI engine that can automatically extract transactions from your receipts (images) and bank statements (PDFs). All data stays local on your device.

🚀 Two Ways to Run AI

Budgetzilla supports two primary ways to run the AI engine:

1. WebGPU (Browser-Native)

The default and most seamless way. It uses the Gemma 4 model directly in your browser’s GPU via the Hugging Face Transformers.js library.

• Prerequisites: A browser with WebGPU support (Chrome 113+, Edge 113+).
• Setup: None! Budgetzilla will automatically download the quantized model weights on first use (approx. 1.2GB).
• Best For: Most users who want a one-click experience.

2. Ollama (Local Server)

For users who prefer to run AI as a background service or on hardware that doesn’t support WebGPU.

• Prerequisites: Ollama installed and running.
• Model: ollama pull gemma4.
• Setup: Configure the Ollama URL and Model Name in Settings.
• Best For: Power users, Linux users, or those with older GPUs.

---

📷 How to Scan Documents

1. Navigate to the Transactions page.
2. Click the Scan Document (AI) button.
3. Choose your input:
   • Image: Upload a photo of a receipt (.jpg, .png).
   • PDF: Upload a bank statement (.pdf).
4. Wait for the AI: The AI will analyze the document and extract:
   • Date: Automatically formatted to YYYY-MM-DD.
   • Merchant: The name of the store or vendor.
   • Amount: The total value found.
   • Category: The AI will attempt to match the merchant to your existing categories.
5. Review & Save: You can edit any extracted field before confirming. Once ready, click Confirm & Save to add the transactions to your ledger.

---

🔒 Privacy & Security

Budgetzilla is local-first, which means your financial data never leaves your machine.

• No Cloud APIs: We don’t use OpenAI, Anthropic, or any other cloud-based LLM.
• No Data Harvesting: Your receipts and statements are processed entirely on your GPU/CPU.
• Offline Ready: Once the model is cached, the AI scanner works even without an internet connection.

---

🛠 Troubleshooting

WebGPU Not Loading

• Ensure you’re using a supported browser (Chrome, Edge, or Arc).
• Check chrome://gpu to verify WebGPU is enabled.
• If it fails, try the Ollama fallback.

Ollama Connection Issues (CORS)

• Verify Ollama is running (ollama list).
• CORS Setup: Browsers block requests to localhost from web apps (like the hosted Vercel version) unless configured otherwise.

macOS:

# Run Ollama with CORS origins allowed
OLLAMA_ORIGINS="https://budgetzilla-app.vercel.app,http://localhost:5173" ollama serve

Windows:

1. Search for “Edit the system environment variables” in the Start menu.
2. Click “Environment Variables”.
3. Add a New User variable:
   • Variable name: OLLAMA_ORIGINS
   • Variable value: https://budgetzilla-app.vercel.app,http://localhost:5173
4. Restart the Ollama application.

• Performance depends on your GPU’s VRAM and compute power.