Document Processor

Go service that watches ~/documents/inbox/ for PDFs and images, uses Kimi K2.5 (via Fireworks API) for OCR and classification, then stores and indexes them.

Features

• File watcher: Monitors inbox for new documents
• OCR + Classification: Kimi K2.5 extracts text and categorizes documents
• Storage: PDFs stored in ~/documents/store/
• Records: Markdown records in ~/documents/records/{category}/
• Index: JSON index at ~/documents/index/master.json
• Expense export: Auto-exports expenses to ~/documents/exports/expenses.csv
• HTTP API: REST endpoints for manual ingestion and search

Setup

1. Set your Fireworks API key:

   export FIREWORKS_API_KEY=your_key_here
   

2. Run the service:

   ./docproc
   

3. Or install as systemd service:

   sudo cp docproc.service /etc/systemd/system/
   # Edit /etc/systemd/system/docproc.service to add your API key
   sudo systemctl daemon-reload
   sudo systemctl enable --now docproc
   

API Endpoints

Endpoint	Method	Description
/health	GET	Health check
/ingest	POST	Upload and process a document (multipart form, field: file)
/search?q=query	GET	Search documents by content
/docs	GET	List all documents
/doc/{id}	GET	Get single document by ID

Directory Structure

~/documents/
├── inbox/      # Drop files here for processing
├── store/      # Processed PDFs stored by hash
├── records/    # Markdown records by category
│   ├── tax/
│   ├── expense/
│   ├── medical/
│   └── ...
├── index/
│   └── master.json  # Document index
└── exports/
    └── expenses.csv  # Expense export

Categories

Documents are classified into:

• tax
• expense
• bill
• invoice
• medical
• receipt
• bank
• insurance
• legal
• correspondence
• other

Usage

Drop a PDF or image into ~/documents/inbox/ and the service will:

1. OCR and classify it
2. Store the original in store/
3. Create a markdown record in records/{category}/
4. Update the master index
5. Export to CSV if it's an expense
6. Delete from inbox

Or POST to /ingest:

curl -X POST http://localhost:9900/ingest -F "file=@receipt.pdf"

Search documents:

curl "http://localhost:9900/search?q=amazon"