# Gemini Cheap Batch — maximum cost stacking: cache + batch + minimal thinking
# Usage: rlmx batch questions.txt --context ./corpus/ --cache --max-cost 2.00
#
# Cost stacking:
#   Base:    $0.075/M input + $0.30/M output (flash-lite)
#   Cache:   90% discount on cached context tokens
#   Batch:   50% discount (when --batch-api is available)
#   Effective: ~$0.0125/M tokens with cache + batch

model:
  provider: google
  model: gemini-3.1-flash-lite-preview
  sub-call-model: gemini-3.1-flash-lite-preview

system: |
  You are a bulk analysis agent optimized for cost efficiency.
  Answer each question concisely using the provided context.
  Do not over-elaborate — aim for the shortest accurate answer.
  {custom_tools_section}

gemini:
  thinking-level: minimal   # Cheapest thinking tier
  media-resolution:
    images: low              # Minimize token usage

criteria: |
  Keep answers under 200 words.
  Reference specific context sources.
  Prioritize precision over verbosity.

tools-level: core   # No batteries needed for simple Q&A

budget:
  max-cost: 2.00
  max-tokens: null
  max-depth: 1

# Cache is always enabled for batch mode
# cache:
#   enabled: true
#   retention: long