config.example.yml

# ragctl Configuration Example
# Copy this file to ~/.ragctl/config.yml or specify with --config flag
#
# Configuration hierarchy (highest priority first):
#   1. CLI flags (--use-llm, --llm-url, etc.)
#   2. Environment variables (ATLAS_USE_LLM, ATLAS_LLM_URL, etc.)
#   3. This config file
#   4. Default values

# ========================================
# LLM Configuration
# ========================================
llm:
  # Enable LLM for text correction and analysis
  use_llm: false

  # Provider: ollama (local), openai (remote), anthropic (remote)
  provider: ollama

  # API URL (auto-detected for each provider if not specified)
  # - Ollama: http://localhost:11434
  # - OpenAI: https://api.openai.com/v1
  # - Anthropic: https://api.anthropic.com/v1
  url: http://localhost:11434

  # API Key (required for remote providers, not needed for local Ollama)
  # Can also use environment variables: OPENAI_API_KEY or ANTHROPIC_API_KEY
  api_key: null

  # Model name
  model: mistral:latest  # For Ollama: mistral:latest, llama3:latest, etc.
  # model: gpt-4o-mini   # For OpenAI
  # model: claude-3-haiku-20240307  # For Anthropic

  # Request timeout (seconds)
  timeout: 60

  # Maximum tokens per request
  max_tokens: 4096

  # Generation temperature (0.0 = deterministic, 2.0 = creative)
  temperature: 0.1

# ========================================
# OCR Configuration
# ========================================
ocr:
  # Enable intelligent OCR routing (auto-selects best engine)
  use_advanced_ocr: false

  # Dictionary ratio threshold (% of real French/English words)
  # < threshold = LOW quality → Qwen-VL advanced OCR
  # ≥ threshold = HIGH quality → Classic OCR (faster)
  # Default: 0.30 (30% real words)
  dictionary_threshold: 0.30

  # Enable dynamic threshold adjustment
  # Adjusts threshold based on:
  #   - Detected language (French: -5%, English: +5%)
  #   - Text length (< 500 chars: +5%, > 2000 chars: -5%)
  dynamic_threshold: true

  # Enable fallback to classic OCR on advanced OCR failures
  enable_fallback: true

  # Qwen-VL (Vision-Language Model) settings
  qwen_vl_url: http://localhost:11434
  qwen_vl_model: qwen/qwen2.5-vl-7b
  qwen_vl_max_tokens: 16384
  qwen_vl_timeout: 120

  # Classic OCR engine (fallback)
  classic_ocr_engine: unstructured  # Options: unstructured, tesseract, pdfminer
  tesseract_languages: fra+eng

# ========================================
# Chunking Configuration
# ========================================
chunking:
  # Chunking strategy
  strategy: semantic  # Options: semantic, sentence, token

  # Maximum tokens per chunk
  max_tokens: 400

  # Token overlap between chunks
  overlap: 50

  # Model for tokenization (used to count tokens)
  model: gpt-3.5-turbo

# ========================================
# Output Configuration
# ========================================
output:
  # Output format
  format: json  # Options: json, jsonl, yaml

  # Include metadata in output
  include_metadata: true

  # Pretty print JSON
  pretty_print: true

  # Generate processing summary (JSON)
  generate_summary: true

# ========================================
# Global Settings
# ========================================

# Log level
log_level: INFO  # Options: DEBUG, INFO, WARNING, ERROR

# ========================================
# Usage Examples
# ========================================
#
# Example 1: Local LLM + Advanced OCR
# --------------------------------------
# llm:
#   use_llm: true
#   provider: ollama
#   url: http://localhost:11434
#   model: mistral:latest
#
# ocr:
#   use_advanced_ocr: true
#   dictionary_threshold: 0.25
#
# chunking:
#   strategy: semantic
#   max_tokens: 500
#
#
# Example 2: Remote OpenAI + Classic OCR
# ----------------------------------------
# llm:
#   use_llm: true
#   provider: openai
#   api_key: sk-...  # Or use OPENAI_API_KEY env var
#   model: gpt-4o-mini
#
# ocr:
#   use_advanced_ocr: false
#   classic_ocr_engine: tesseract
#
#
# Example 3: No LLM + Advanced OCR (Recommended for scanned docs)
# -----------------------------------------------------------------
# llm:
#   use_llm: false
#
# ocr:
#   use_advanced_ocr: true
#   dictionary_threshold: 0.30
#   dynamic_threshold: true
#
# chunking:
#   strategy: semantic
#   max_tokens: 400
#
# ========================================
# Environment Variables (Alternative to this file)
# ========================================
#
# All settings can be set via environment variables:
#
# ATLAS_USE_LLM=true
# ATLAS_LLM_PROVIDER=ollama
# ATLAS_LLM_URL=http://localhost:11434
# ATLAS_LLM_MODEL=mistral:latest
#
# ATLAS_USE_ADVANCED_OCR=true
# ATLAS_OCR_DICTIONARY_THRESHOLD=0.30
# ATLAS_OCR_DYNAMIC_THRESHOLD=true
#
# ATLAS_CHUNK_STRATEGY=semantic
# ATLAS_CHUNK_MAX_TOKENS=400
# ATLAS_CHUNK_OVERLAP=50
#
# ATLAS_LOG_LEVEL=INFO
#
# ========================================
# CLI Flags (Highest Priority)
# ========================================
#
# All settings can be overridden via CLI flags:
#
# ragctl chunk document.pdf \
#   --use-llm \
#   --llm-url http://localhost:11434 \
#   --llm-model mistral:latest \
#   --use-advanced-ocr \
#   --ocr-threshold 0.25 \
#   --max-tokens 500 \
#   --show