os-image-composer.yml

# OS Image Composer - Global Configuration
# This file contains tool-level settings that apply across all image builds.
# Image-specific parameters should be defined in the image specification.

# Core tool settings
workers: 8
# Number of concurrent download workers (1-100, default: 8)
# Higher values speed up package downloads but consume more network/CPU resources
# Recommended: 8-16 for most systems, 20+ for high-bandwidth servers

config_dir: "./config"
# Directory containing configuration files for different target OSs (default: ./config)
# Should contain subdirectories for general and each target OS config files.

cache_dir: "./cache"
# Package cache directory where downloaded RPMs/DEBs are stored (default: ./cache)
# This directory persists between builds for package reuse
# Should have sufficient space (typically 1-5GB depending on image size)

work_dir: "./workspace"
# Working directory for build operations and image assembly (default: ./workspace)
# Contains temporary build artifacts, extracted packages, and final images
# Hosts the per-provider chrootenv/chrootbuild trees used for entering/exiting chroot
# Requires substantial space during builds (2-10GB typical)

temp_dir: "./tmp"
# Temporary directory for short-lived files like GPG keys and metadata parsing
# Empty value uses system default (/tmp on Linux, %TEMP% on Windows)
# Used for: GPG verification files, decompressed metadata, parsing operations
# Files here are deleted within seconds/minutes of creation

# Logging configuration
logging:
  level: "info"
  # Log verbosity level (default: info)
  # - debug: Most verbose, shows all operations and data structures
  # - info:  Normal output, shows progress and important events
  # - warn:  Only warnings and errors, minimal output
  # - error: Only errors, very quiet operation
  file: "os-image-composer.log"
  # Tee logs to this file in addition to stdout/stderr (overwritten on each run)

# AI-powered template generation configuration (optional)
# All settings have sensible defaults - this section can be omitted entirely
ai:
  # Provider for embeddings and chat completions
  provider: "ollama"
  # - ollama: Local inference (default, requires Ollama running on localhost:11434)
  # - openai: Cloud inference (requires OPENAI_API_KEY environment variable)

  # Directory containing template files for RAG knowledge base
  templates_dir: "./image-templates"

  # Ollama-specific settings (used when provider: ollama)
  ollama:
    base_url: "http://localhost:11434"
    embedding_model: "nomic-embed-text"
    # Embedding models: nomic-embed-text (768 dims), mxbai-embed-large (1024 dims)
    chat_model: "llama3.2"
    # Chat models: llama3.2, mistral, codellama, etc.
    timeout: "120s"

  # OpenAI-specific settings (used when provider: openai)
  # openai:
  #   embedding_model: "text-embedding-3-small"
  #   chat_model: "gpt-4o-mini"
  #   timeout: "60s"

  # Embedding cache configuration
  cache:
    enabled: true
    dir: "./cache/ai-embeddings"
    # Cache stores embeddings to avoid re-computing on each run

  # Hybrid scoring weights for template retrieval
  # scoring:
  #   semantic_weight: 0.70   # Embedding similarity (0.0-1.0)
  #   keyword_weight: 0.20    # Keyword matching (0.0-1.0)
  #   package_weight: 0.10    # Package matching (0.0-1.0)