example_app_config.env

# Rename this file to app_config.env and place it in the folder config/ (i.e. it will be located at app_base_folder/config/app_config.env). The app will then automatically load in these variables at startup. See tools/config.py for all the possible config variables you can set, or src/app_settings.qmd for descriptions. Below are some suggested config variables to start

# General app run options
TESSERACT_FOLDER=tesseract/ # If in a custom folder, not needed if in PATH
POPPLER_FOLDER=poppler/poppler-24.02.0/Library/bin/ # If in a custom folder, Not needed if in PATH

GRADIO_SERVER_NAME=127.0.0.1
GRADIO_SERVER_PORT=7860

USER_GUIDE_URL=<ENTER_URL>
CUSTOM_BOX_COLOUR=(128, 128, 128)
RUN_FASTAPI=False
FAVICON_PATH=favicon.png
INTRO_TEXT=intros/short_intro.txt

# GUI options
SHOW_QUICKSTART=False
SHOW_SUMMARISATION=True
SHOW_EXAMPLES=True
SHOW_DIFFICULT_OCR_EXAMPLES=True
SHOW_LANGUAGE_SELECTION=True
SHOW_WHOLE_DOCUMENT_TEXTRACT_CALL_OPTIONS=False
SHOW_COSTS=True
SHOW_LOCAL_OCR_MODEL_OPTIONS=True
SHOW_ALL_OUTPUTS_IN_OUTPUT_FOLDER=True
SHOW_PII_IDENTIFICATION_OPTIONS=True
SHOW_LOCAL_PII_DETECTION_OPTIONS=True
SHOW_OCR_GUI_OPTIONS=True
EXTRACTION_AND_PII_OPTIONS_OPEN_BY_DEFAULT=True

# Model / redaction process options
DEFAULT_LOCAL_OCR_MODEL=tesseract
OVERWRITE_EXISTING_OCR_RESULTS=False
PREPROCESS_LOCAL_OCR_IMAGES=False # Whether to apply corrections to input images before processing. Will slow down redaction processes
MAX_WORKERS=4 # How many workers should be working in parallel to run various text extraction/redaction tasks. Adjust depending on how many CPUs your computer has

EFFICIENT_OCR=True
OVERWRITE_EXISTING_OCR_RESULTS=True
INCLUDE_OCR_VISUALISATION_IN_OUTPUT_FILES=True

# Redaction box appearance
CUSTOM_BOX_COLOUR=(128, 128, 128)
USE_GUI_BOX_COLOURS_FOR_OUTPUTS=False

# Image save options
SAVE_PAGE_OCR_VISUALISATIONS=True
SAVE_PREPROCESS_IMAGES=True

# Saving and logging variables
SAVE_LOGS_TO_CSV=True
SESSION_OUTPUT_FOLDER=True # Save outputs into user session folders
DISPLAY_FILE_NAMES_IN_LOGS=False

# PaddleOCR
SHOW_PADDLE_MODEL_OPTIONS=False
LOAD_PADDLE_AT_STARTUP=False
PADDLE_MAX_WORKERS=4 # Number of simultaneous workers for Paddle OCR tasks. Generally advised to keep at 1, but may work with 2 or more depending on your system.

# GUI show VLM/LLM models
SHOW_HYBRID_MODELS=False
SHOW_CUSTOM_VLM_ENTITIES=False
SHOW_VLM_MODEL_OPTIONS=True
SHOW_INFERENCE_SERVER_PII_OPTIONS=False
SHOW_INFERENCE_SERVER_VLM_OPTIONS=False
SHOW_TRANSFORMERS_LLM_PII_DETECTION_OPTIONS=False

# VLM using Transformers options
SELECTED_LOCAL_TRANSFORMERS_VLM_MODEL=Qwen3.5-9B
QUANTISE_VLM_MODELS=False
USE_TRANSFORMERS_VLM_MODEL_AS_LLM=True
LOCAL_TRANSFORMERS_LLM_PII_MODEL_CHOICE=None
QUANTISE_TRANSFORMERS_LLM_MODELS=False
LOAD_TRANSFORMERS_LLM_PII_MODEL_AT_START=False
LOAD_TRANSFORMERS_VLM_MODEL_AT_START=True

# VLM using inference server options (vLLM / Llama.cpp server)
INFERENCE_SERVER_API_URL=http://192.168.0.220:8080
USE_LLAMA_SWAP=True
INFERENCE_SERVER_LLM_PII_MODEL_CHOICE=qwen_3_5_27b

# General VLM / LLM options
VLM_DISABLE_QWEN3_5_THINKING=True
LLM_MAX_NEW_TOKENS=8192
CUSTOM_VLM_BACKEND=bedrock_vlm # Which model type to use to do face / signature detection. Can choose from "transformers_vlm", "inference_vlm", "bedrock_vlm"

# AWS related variables
RUN_AWS_FUNCTIONS=True # Set to False if you don't want to run AWS functions. You can remove all the environment variables in the following section if you don't want to use them
AWS_REGION=example-region
DOCUMENT_REDACTION_BUCKET=example-bucket

SHOW_AWS_TEXT_EXTRACTION_OPTIONS=True
SHOW_AWS_PII_DETECTION_OPTIONS=True

SHOW_AWS_EXAMPLES=True
RUN_ALL_EXAMPLES_THROUGH_AWS=True

SAVE_LOGS_TO_DYNAMODB=True
ACCESS_LOG_DYNAMODB_TABLE_NAME=example-dynamodb-access-log
USAGE_LOG_DYNAMODB_TABLE_NAME=example-dynamodb-usage
FEEDBACK_LOG_DYNAMODB_TABLE_NAME=example-dynamodb-feedback

# AWS Textract options
SHOW_WHOLE_DOCUMENT_TEXTRACT_CALL_OPTIONS=True
LOAD_PREVIOUS_TEXTRACT_JOBS_S3=True
TEXTRACT_WHOLE_DOCUMENT_ANALYSIS_BUCKET=example-bucket-output
INCLUDE_FORM_EXTRACTION_TEXTRACT_OPTION=False
INCLUDE_LAYOUT_EXTRACTION_TEXTRACT_OPTION=False
INCLUDE_TABLE_EXTRACTION_TEXTRACT_OPTION=False
INCLUDE_FACE_IDENTIFICATION_TEXTRACT_OPTION=False # Needs a VLM option available to work

# AWS VLM / LLM options
SHOW_BEDROCK_VLM_MODELS=False
SHOW_AWS_BEDROCK_LLM_MODELS=False
HYBRID_TEXTRACT_BEDROCK_VLM=False

CLOUD_LLM_PII_MODEL_CHOICE=amazon.nova-pro-v1:0
CLOUD_LLM_PII_CUSTOM_INSTRUCTIONS_MODEL_CHOICE=anthropic.claude-sonnet-4-6 #amazon.nova-pro-v1:0
CLOUD_VLM_MODEL_CHOICE=amazon.nova-pro-v1:0 # other possibles: anthropic.claude-sonnet-4-6 #qwen.qwen3-vl-235b-a22b # anthropic.claude-sonnet-4-6 #
CLOUD_SUMMARISATION_MODEL_CHOICE=amazon.nova-lite-v1:0

# Cost code related variables
SHOW_COSTS=True
GET_COST_CODES=True
COST_CODES_PATH=config/cost_codes.csv
ENFORCE_COST_CODES=True
DEFAULT_COST_CODE=example_cost_code

# S3 cost codes
S3_COST_CODES_PATH=cost_codes.csv