document-parser / README.md
segovia
multitenant supprt
028d116
|
Raw History Blame Contribute Delete
3.3 kB
---
title: Invoice Parser Agent
emoji: 📚
colorFrom: pink
colorTo: purple
sdk: gradio
sdk_version: 6.20.0
python_version: '3.12'
app_file: app.py
pinned: false
---
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
## Drive Folder Scan (Unprocessed Files)
The app supports scanning one or more Google Drive folders and parsing only files that were not processed before.
### Required environment variables
- `GOOGLE_SERVICE_ACCOUNT_JSON` or `GOOGLE_SERVICE_ACCOUNT_FILE`
- `GOOGLE_DRIVE_FOLDER_IDS`: comma-separated Drive folder IDs
Optional:
- `FIREBASE_TRACK_PROCESSED`: defaults to `true`
- `FIREBASE_TRACKER_COLLECTION`: defaults to `processed_files`
- `FIREBASE_DOCUMENTS_COLLECTION`: defaults to `processed_documents`
- `GOOGLE_DRIVE_FACTURAS_BASE_PATH`: defaults to `Facturas`
- `GOOGLE_DRIVE_INVOICE_ROOT_FOLDER_ID`: optional Drive folder id used as root when creating `GOOGLE_DRIVE_FACTURAS_BASE_PATH`
### How it works
1. Click **Scan Drive folders** in the UI.
2. The app recursively scans configured Drive folders.
3. Supported file types are downloaded to a temporary local folder (`.txt`, `.pdf`, `.docx`, `.jpg`, `.jpeg`, `.png`).
4. Files already registered in Firebase are skipped.
5. New files are parsed and appended to the same grid used by uploads.
## Drive Orchestration (Status + modificationDate)
The orchestration script scans Drive, checks if a document was already processed, parses only new/modified documents, and saves parsed payload + lifecycle status in Firebase.
Before each document is persisted, the file is moved in Drive using this path rule:
- `Facturas/yyyyMM` for valid invoices where `fecha` can be parsed.
- `Facturas/Corrupted` when required sheet data is missing or `fecha` is not parseable.
You can change `Facturas` with `GOOGLE_DRIVE_FACTURAS_BASE_PATH`.
### Status rules
- `Parsed`: document was never processed before.
- `Modified`: document exists in Firebase but Drive `modifiedTime` changed.
- `Sent`: document was parsed and then sent to Google Sheets.
### Run orchestration
```bash
python document_orchester.py --model gpt-4o
```
Send parsed rows to Google Sheets and mark them as `Sent`:
```bash
python document_orchester.py --model gpt-4o --send
```
Optional flags:
- `--no-subfolders`: scan only direct files in configured folders.
- `--output <path>`: save orchestration summary JSON to a file.
## Public API Endpoint For Orchestrator
The Gradio app exposes a public API endpoint that internally calls the orchestrator flow.
Security:
- Set `ORCHESTRATOR_API_KEY` in the server environment.
- Every endpoint call must include that key in the payload.
Endpoint name:
- `orchestrate_drive_documents`
Local URL example:
```bash
curl -X POST "http://localhost:7860/gradio_api/call/orchestrate_drive_documents" \
-H "Content-Type: application/json" \
-d '{"data": ["gpt-4o", true, false, "YOUR_API_KEY"]}'
```
`data` argument order:
1. `model` (string)
2. `include_subfolders` (boolean)
3. `send_to_sheet` (boolean)
4. `api_key` (string)
Response payload is the same summary produced by `orchestrate_drive_documents` in `document_orchester.py`.
When a row is missing required sheet data, it is written to a worksheet named `corrupted data` instead of the main sheet.