config.example.yml

target:
  baseUrl: https://api.unorouter.com
  systemAccessToken: target-system-access-token
  userId: 1
blacklist:
  - kiro
  - nsfw
modelMapping:
  claude-sonnet-4-5-20250929-thinking: claude-sonnet-4-5-20250929
# Per-user, per-model rate limits for the big FREE models that get swarmed.
# Globs match the published `:free` names only; paid models are never limited.
# success = successful requests per user per gateway window. Optional total =
# attempts including failures (absent = unlimited, a failed request never
# burns the budget).
rateLimit:
  models:
    kimi-*:free: { success: 20 }
    glm-*:free: { success: 15 }
    deepseek-*:free: { success: 10 }
providers:
  # ── new-api upstream ──
  - type: newapi
    name: upstream-main
    baseUrl: https://www.newapi.ai
    systemAccessToken: upstream-system-access-token
    userId: 1
    enabledVendors:
      - anthropic
      - openai
    enabledModels:
      - claude-*-4-5*
      - gpt-5*
    priceAdjustment: -0.67

  # ── sub2api ──
  - type: sub2api
    name: sub2api-fallback
    baseUrl: https://sub2api.unorouter.com
    adminApiKey: admin-api-key
  # disabled providers (uncomment to enable)
  # - type: openrouter
  #   name: openrouter
  #   apiKey: sk-or-...
  # - type: nvidia
  #   name: nvidia-nim
  #   apiKey: nvapi-...
  # Free providers with tight RPM/daily caps set acceptRateLimited so a 429 during
  # probing keeps the model (capacity throttle, not breakage). It is emitted disabled
  # and new-api's auto-test re-enables it once the limit clears. Only set this where a
  # 429 means "quota spent", never "model broken".
  # - type: cloudflare
  #   name: cf1
  #   apiKey: cf-...
  #   acceptRateLimited: true