I Built a Local AI Office Assistant That Never Sends Your Documents to the Cloud
Most AI document tools today have one thing in common:
Your files leave your computer.
Invoices, contracts, receipts, meeting notes, customer documents — everything gets uploaded somewhere.
I wanted something different.
So I built a local AI-powered Office Assistant for Windows that runs directly on your machine using Gemma 4 through Ollama.
No subscriptions. No API keys. No cloud processing.
Why I Built It
I kept running into repetitive document tasks such as:
- Summarizing PDFs
- Extracting information from receipts
- Converting one document type into another
- Pulling action items from meeting notes
- Creating invoices from timesheets
The process was repetitive and often involved copying sensitive files into online tools.
I wanted a system where files stay on the user's machine.
What the App Can Do
📄 Bill Summation
Upload:
- Receipt photos
- Scanned invoices
- JPG/PNG files
- Excel spreadsheets
The AI reads:
- Vendor names
- Dates
- Line items
- Totals
Once processing is complete:
✅ Review extracted information
✅ Correct any data if needed
✅ Save formatted Excel files
✅ Generate a Master Total spreadsheet with linked bills and totals
Useful for:
- Expense reports
- Monthly accounting
- Supplier reconciliation
💬 Ask My Document
Load almost any document:
- Word
- Excel
- Text files
- Images
Then ask questions in plain English:
Examples:
Give me a 100-word summary
Extract all phone numbers
Draft a professional response
List all action items
The original document is never modified.
✨ Transform Documents
Convert documents into entirely different formats:
Examples:
- Timesheet → Invoice
- CV → Cover Letter
- Meeting Notes → Structured Minutes
- Letter → Professional Reply
Supports:
- Custom instructions
- Template-based formatting
- Single file processing
- Batch folder processing
Features
- Completely local processing
- No cloud upload required
- Gemma 4 AI via Ollama
- Batch processing
- Email notifications
- Configurable output folders
- Custom templates
- Debug logs
- Professional Excel outputs
- PDF support
Things I Learned Building It
Building local AI turned out to be very different from using cloud APIs.
Challenges included:
- OCR accuracy
- Memory management
- Model selection
- Performance optimization
- Handling large document workflows
A GPU definitely helps, but smaller Gemma models can still work on modest hardware.
What Makes It Different
Many AI tools focus on convenience.
I wanted to focus on privacy too.
This means:
- Documents never leave your machine
- No recurring subscriptions
- No API keys
- No external servers involved
I think local AI is going to become a larger part of everyday productivity workflows.
Interested?
I'm currently looking for feedback and early users.
If you'd like to try it, leave a comment or send me a message and I'll send you a download link.
I'd also love to know:
What repetitive document task would you automate first?
Thanks for reading.