config/app_config.env.example

# Main application settings (copy to config/app_config.env — do not commit secrets).
#
# Loaded at startup by tools/config.py (APP_CONFIG_PATH defaults to config/app_config.env).
# Variables already set in the process environment (e.g. Docker compose environment:)
# are not overridden by this file.
#
# Full list of supported keys: tools/config.py and the user guide.
# For Docker Compose-only overrides on redaction-app services, see docker_app_config.env.example.
# Optional separate AWS keys file: set AWS_CONFIG_PATH=config/aws_config.env

# --- Local dependencies (Windows/Linux paths if not on PATH) ---
TESSERACT_FOLDER=tesseract/
POPPLER_FOLDER=poppler/poppler-24.02.0/Library/bin/
# TESSERACT_DATA_FOLDER=

# --- Input / output folders ---
# GRADIO_INPUT_FOLDER=input/
# GRADIO_OUTPUT_FOLDER=output/
SESSION_OUTPUT_FOLDER=True

# --- Gradio / FastAPI ---
GRADIO_SERVER_NAME=127.0.0.1
GRADIO_SERVER_PORT=7860
RUN_FASTAPI=False
# RUN_MCP_SERVER=False
# ROOT_PATH=
# FASTAPI_ROOT_PATH=/
# ALLOWED_HOSTS=
# MAX_DOC_PAGES=3000
# MAX_FILE_SIZE=250mb
# AGENT_API_KEY=

# --- Intro / branding ---
USER_GUIDE_URL=<ENTER_URL>
FAVICON_PATH=favicon.png
INTRO_TEXT=intros/short_intro.txt
# PI agent Gradio UI (agent-redact/pi/gradio_app.py) — separate from INTRO_TEXT above
# AGENT_INTRO_TEXT=intros/pi_intro.txt

# --- GUI visibility ---
SHOW_QUICKSTART=False
SHOW_SUMMARISATION=True
SHOW_EXAMPLES=True
SHOW_DIFFICULT_OCR_EXAMPLES=True
SHOW_LANGUAGE_SELECTION=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
SHOW_LOCAL_OCR_MODEL_OPTIONS=True
SHOW_AWS_TEXT_EXTRACTION_OPTIONS=True
SHOW_AWS_PII_DETECTION_OPTIONS=True
SHOW_AWS_EXAMPLES=True
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
SHOW_BEDROCK_VLM_MODELS=False
SHOW_AWS_BEDROCK_LLM_MODELS=False

# --- OCR / redaction processing ---
DEFAULT_LOCAL_OCR_MODEL=tesseract
OVERWRITE_EXISTING_OCR_RESULTS=False
PREPROCESS_LOCAL_OCR_IMAGES=False
MAX_WORKERS=4
EFFICIENT_OCR=True
INCLUDE_OCR_VISUALISATION_IN_OUTPUT_FILES=True
# Local OCR reading order: column | legacy | paddle_native
LOCAL_OCR_READING_ORDER=column
CONVERT_LINE_TO_WORD_LEVEL=True
MERGE_BOUNDING_BOXES=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

# --- PaddleOCR ---
SHOW_PADDLE_MODEL_OPTIONS=False
LOAD_PADDLE_AT_STARTUP=False  # On ZeroGPU, Paddle loads in the GPU worker on first OCR (not at Space startup).
PADDLE_DEVICE=gpu:0  # transformers backend device (cpu, gpu, gpu:0). Auto-defaults to gpu:0 on ZeroGPU.
PADDLE_MAX_WORKERS=4

# --- VLM / LLM — local Transformers ---
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 / LLM — inference server (vLLM / llama.cpp) ---
INFERENCE_SERVER_API_URL=http://192.168.0.220:8080
USE_LLAMA_SWAP=True
INFERENCE_SERVER_LLM_PII_MODEL_CHOICE=qwen_3_5_27b

# --- VLM / LLM — general / cloud backend ---
VLM_DISABLE_QWEN3_5_THINKING=True
LLM_MAX_NEW_TOKENS=8192
CUSTOM_VLM_BACKEND=bedrock_vlm
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
CLOUD_VLM_MODEL_CHOICE=amazon.nova-pro-v1:0
CLOUD_SUMMARISATION_MODEL_CHOICE=amazon.nova-lite-v1:0

# --- AWS (core) ---
RUN_AWS_FUNCTIONS=True
PRIORITISE_SSO_OVER_AWS_ENV_ACCESS_KEYS=True
AWS_REGION=example-region
# AWS_ACCESS_KEY=
# AWS_SECRET_KEY=
DOCUMENT_REDACTION_BUCKET=example-bucket
# S3_OUTPUTS_BUCKET=
# S3_OUTPUTS_FOLDER=
# SAVE_OUTPUTS_TO_S3=False
RUN_ALL_EXAMPLES_THROUGH_AWS=True

# --- AWS Textract ---
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

# --- Logging ---
SAVE_LOGS_TO_CSV=True
SAVE_LOGS_TO_DYNAMODB=True
DISPLAY_FILE_NAMES_IN_LOGS=False
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

# --- Cost codes ---
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_PATH=cost_codes.csv

# --- Cognito / hosted auth (when RUN_FASTAPI=True) ---
# COGNITO_AUTH=False
# AWS_USER_POOL_ID=
# AWS_CLIENT_ID=
# AWS_CLIENT_SECRET=

# --- Post-redaction Pass 1 QA (pre-review_apply sanity checks) ---
# POST_REDACT_PASS1_QA=False
# POST_REDACT_PASS1_AUTO_PRUNE=False
# POST_REDACT_PASS1_USE_DENY_ALLOW_LISTS=True
# POST_REDACT_PASS1_MUST_REDACT_PATH=
# POST_REDACT_PASS1_MUST_NOT_REDACT_PATH=