# ============================================================================
# PulseForge AI - Production LiteLLM Cascade & KV-Prompt-Cache Router (1 of 9)
# Target Repository: langgenius/dify
# Primary Frontier Model: Claude Sonnet 4.6 (Anthropic) (anthropic.claude-sonnet-4-6)
# Raw Rates: $3.0/1M in | $0.3/1M KV-read | $15.0/1M out
# Free Starter CLAUDE.md ($0): https://exactbaseline.com/api/launch-kit/free-claude-md?repo=langgenius/dify
# 1-Click $4.99 Drop-In Config Pack (3 Core Files): https://exactbaseline.com/buy/micro_pass?repo=langgenius/dify&source=yaml_export
# 1-Click $9.99 Complete 9-File Production .ZIP: https://exactbaseline.com/buy/repo_pass_999?repo=langgenius/dify&source=yaml_export_999
# Upgrade to Full Hardening & Launch Kit ($49): https://exactbaseline.com/buy/instant_boost?repo=langgenius/dify
# ============================================================================
model_list:
  - model_name: frontier-architect
    litellm_params:
      model: anthropic.claude-sonnet-4-6
      cache_control_injection_points:
        - location: message
          role: system
      max_tokens: 4096

  - model_name: fast-diff-worker
    litellm_params:
      model: deepseek/deepseek-reasoner
      max_tokens: 4096

  - model_name: local-zero-cost
    litellm_params:
      model: ollama/qwen2.5-coder:32b
      api_base: http://localhost:11434

router_settings:
  routing_strategy: cost-based-routing
  enable_pre_call_checks: true
  num_retries: 2
  fallbacks:
    - fast-diff-worker: ["frontier-architect"]
    - local-zero-cost: ["fast-diff-worker"]

litellm_settings:
  cache: true
  cache_params:
    type: redis-semantic
    similarity_threshold: 0.92
    ttl: 300