# ContextWeave CLI config example.
#
# This file is intentionally verbose. It is meant to be a single AI-readable
# reference for generating ContextWeave CLI configs, close to what the Setup Wizard
# creates for a Balanced, DeepSeek-only workflow profile.
#
# contextweave-cli resolves config in this order:
# 1. --config PATH
# 2. CONTEXTWEAVE_CONFIG
# 3. nearest contextweave.yaml, contextweave.yml, .contextweave.yaml, or .contextweave.yml walking upward from cwd
# 4. the platform default shown by `contextweave-cli config path`
#
# The schema mirrors the app setup UI:
# - connections.* is close to ConnectionDraft
# - workflow_profiles.* is close to WorkflowProfileDetail
# - workflow_profiles.*.routes[] is close to WorkflowStepRoute

version: 1

# The workflow profile key under workflow_profiles to use for new CLI books.
default_workflow_profile: balanced_deepseek

connections:
  # Balanced DeepSeek profile uses Flash for cheaper, lower-latency context work.
  deepseek_flash:
    # Human-readable name. The mapping key, deepseek_flash, is what routes use as connection_id.
    display_name: DeepSeek V4 Flash

    # Supported values include: deepseek, gemini, openai, anthropic, openai_compatible.
    provider: deepseek

    # Prefer api_key_env so secrets stay out of config files, book configs, and task snapshots.
    # The environment variable must be present when running `contextweave-cli config validate` or `contextweave-cli run`.
    api_key_env: DEEPSEEK_API_KEY

    # api_key is accepted for private/local configs, but do not set both api_key and api_key_env.
    # api_key: sk-...

    base_url: https://api.deepseek.com
    default_model: deepseek-v4-flash

    # These defaults match the Setup Wizard template for DeepSeek V4 Flash.
    temperature: 0
    timeout: 180
    max_retries: 3
    concurrency: 2500

  # Balanced DeepSeek profile uses Pro for quality-sensitive generation/review work.
  deepseek_pro:
    display_name: DeepSeek V4 Pro
    provider: deepseek
    api_key_env: DEEPSEEK_API_KEY
    base_url: https://api.deepseek.com
    default_model: deepseek-v4-pro

    # These defaults match the Setup Wizard template for DeepSeek V4 Pro.
    temperature: 0
    timeout: 300
    max_retries: 3
    concurrency: 500

workflow_profiles:
  balanced_deepseek:
    profile_id: balanced_deepseek
    name: Balanced DeepSeek
    kind: shared

    # The language ContextWeave should translate into for books created from this profile.
    target_language: English

    routes:
      # Extractor builds initial term candidates from source text.
      # Balanced DeepSeek uses V4 Flash here because extraction is high-volume and
      # benefits from speed/cost efficiency more than maximum prose quality.
      - step_id: extractor
        step_label: Extractor
        connection_id: deepseek_flash
        model: deepseek-v4-flash
        step_config:
          # Number of extra glossary-extraction passes. The wizard default is 1.
          max_gleaning: 1

          # The Setup Wizard enables DeepSeek V4 thinking for selected DeepSeek V4 routes.
          kwargs:
            extra_body:
              thinking:
                type: enabled

      # Summarizer condenses term/context information used later by translation.
      # Balanced DeepSeek uses V4 Flash because the output is short support context,
      # not the final translated prose.
      - step_id: summarizer
        step_label: Summarizer
        connection_id: deepseek_flash
        model: deepseek-v4-flash
        step_config:
          kwargs:
            extra_body:
              thinking:
                type: enabled

      # Glossary translator turns source terms into target-language terms.
      # Balanced DeepSeek uses V4 Pro because glossary mistakes propagate through
      # the whole book, so quality matters more than the small extra cost here.
      - step_id: glossary_translator
        step_label: Glossary translator
        connection_id: deepseek_pro
        model: deepseek-v4-pro
        step_config:
          kwargs:
            extra_body:
              thinking:
                type: enabled

      # Translator performs the main chunk translation.
      # Balanced DeepSeek uses V4 Pro for better style, consistency, and reasoning.
      - step_id: translator
        step_label: Translator
        connection_id: deepseek_pro
        model: deepseek-v4-pro
        step_config:
          # DeepSeek wizard defaults for text translation chunking.
          chunk_size: 1000
          max_tokens_per_llm_call: 3500

          # EPUB only: strip ruby annotation markup before translation.
          strip_epub_ruby: true

          kwargs:
            extra_body:
              thinking:
                type: enabled

      # Polish improves readability and style after the main translation pass.
      # Balanced DeepSeek uses V4 Pro because this step edits final user-visible prose.
      - step_id: polish
        step_label: Polish
        connection_id: deepseek_pro
        model: deepseek-v4-pro
        step_config:
          kwargs:
            extra_body:
              thinking:
                type: enabled

      # Reviewer checks quality and catches issues after translation/polish.
      # Balanced DeepSeek uses V4 Pro because review is reasoning-heavy and should
      # prefer precision over speed.
      - step_id: reviewer
        step_label: Reviewer
        connection_id: deepseek_pro
        model: deepseek-v4-pro
        step_config:
          kwargs:
            extra_body:
              thinking:
                type: enabled

      # DeepSeek-only Balanced does not include OCR, image reembedding, or manga
      # image translation routes because those need vision/image-capable providers.
      # Add Gemini/OpenAI routes for these steps if the book needs scanned PDF,
      # scanned-book, manga, or image text handling.