.env.sample

# PDF processing backend: auto (default), pdf_oxide, pymupdf4llm, pypdf2, gemini
# auto: detects text-based vs scanned PDFs; text-based -> pdf_oxide, scanned -> Gemini OCR
PDF_BACKEND=auto

# Gemini model used for scanned PDF OCR
GEMINI_MODEL_FOR_VISION=gemini-2.0-flash-001

# --- PDF Text-vs-Scanned Detection (adaptive sqrt scaling) ---
# Minimum number of pages to sample
PDF_SAMPLE_MIN=2
# Maximum number of pages to sample (even for very large PDFs)
PDF_SAMPLE_MAX=20
# Minimum words per page to consider it text-based
PDF_WORD_THRESHOLD=75
# Ratio of sampled pages that must exceed word threshold to classify as text-based
PDF_TEXT_RATIO=0.7
# Google Cloud credentials (Required for Gemini client)
GOOGLE_CLOUD_PROJECT=

# it should be us-central1 for Batch Predictions with Gemini
GOOGLE_CLOUD_LOCATION=us-central1

# this should be in us-central1 for Batch Predictions with Gemini.
# Example: GCS_BUCKET=gs://your-bucket-name
GCS_BUCKET=

# Path to Google Cloud service account JSON file (it should have the proper permissions for GCS and Gemini)
# Example: GOOGLE_APPLICATION_CREDENTIALS=/path/to/your/service-account-file.json
GOOGLE_APPLICATION_CREDENTIALS=/app/google-application-credentials.json

# duckdb file for tracking batch operations
DUCKDB_FILE=batch_tasks.duckdb

# Maximum file size for uploads in Megabytes (Optional)
# Example: MAX_FILE_SIZE_MB=50
MAX_FILE_SIZE_MB=

# API Authentication Bearer Token (Optional. If set, all endpoints will require this token)
# Example: API_AUTH_TOKEN=your-secret-bearer-token
API_AUTH_TOKEN=