config.yaml

auto_update:
  allow_major: false
  apply_command: faigate-update
  enabled: false
  maintenance_window:
    days:
    - sat
    - sun
    enabled: false
    end_hour: 5
    start_hour: 2
    timezone: UTC
  max_unhealthy_providers: 0
  min_release_age_hours: 0
  provider_scope:
    allow_providers: []
    deny_providers:
    - openrouter-fallback
  require_healthy_providers: true
  rollout_ring: early
  verification:
    command: faigate-health
    enabled: false
    rollback_command: ''
    timeout_seconds: 30
metadata:
  enabled: true
  on_startup: false
  refresh_interval_hours: 24
  timeout_seconds: 10
client_profiles:
  default: generic
  enabled: true
  presets:
  - openclaw
  - n8n
  - cli
  profiles:
    cli:
      routing_mode: auto
    generic: {}
    local-only:
      capability_values:
        local: true
    n8n:
      prefer_tiers:
      - cheap
      - default
      routing_mode: eco
    antigravity:
      prefer_providers:
      - gemini-flash-lite
      - gemini-flash
      - gemini-pro-high
      prefer_tiers:
      - cheap
      - default
      routing_mode: eco
    claude:
      routing_mode: coding-auto
    codex:
      routing_mode: coding-auto
    deepseek-cli:
      prefer_providers:
      - deepseek-v4-flash
      - deepseek-v4-pro
      - anthropic-haiku
      prefer_tiers:
      - default
      - reasoning
      routing_mode: auto
    gemini-cli:
      prefer_providers:
      - gemini-flash
      - gemini-flash-lite
      - gemini-pro-high
      prefer_tiers:
      - cheap
      - default
      - mid
      routing_mode: auto
    kilocode:
      routing_mode: coding-fast
    openclaw:
      routing_mode: coding-auto
    opencode:
      # The chatgpt.com Codex backend gates `gpt-5-codex` to Pro accounts only,
      # so the auto-router must not pick codex providers as a fallback for the
      # opencode client. Explicit invocation via model_shortcuts still works.
      deny_providers:
      - openai-codex
      - openai-codex-mini
      - openai-codex-spark
      - openai-codex-high
      - openai-codex-xhigh
      - openai-codex-low
      - openai-codex-5.4-low
      - openai-codex-5.4-medium
      - openai-codex-5.4-high
      - openai-codex-5.4-xhigh
      routing_mode: coding-auto
    # ── opencode named profiles (selectable as model in opencode UI) ─────────
    coding:
      deny_providers:
      - openai-codex
      - openai-codex-mini
      - openai-codex-spark
      - openai-codex-high
      - openai-codex-xhigh
      - openai-codex-low
      - openai-codex-5.4-low
      - openai-codex-5.4-medium
      - openai-codex-5.4-high
      - openai-codex-5.4-xhigh
      routing_mode: coding-auto
    coding-free:
      prefer_providers:
      - pollinations
      - groq
      - cerebras
      - kilocode
      - blackbox-free
      - gemini-flash-lite
      prefer_tiers:
      - free
      - cheap
      routing_mode: free
    research:
      routing_mode: premium
  rules:
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - antigravity
    profile: antigravity
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - claude
          - claude-code
    profile: claude
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - codex
    profile: codex
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - deepseek-cli
    profile: deepseek-cli
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - gemini-cli
    profile: gemini-cli
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - kilocode
          - kilo
    profile: kilocode
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - opencode
    profile: opencode
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - coding
          - coding-auto
    profile: coding
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - coding-free
          - free
    profile: coding-free
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - research
    profile: research
fallback_chain:
- deepseek-v4-flash
- anthropic-haiku
- gemini-flash
- deepseek-v4-pro
- anthropic-sonnet
- gemini-pro-high
- openai-gpt4o
- openrouter-fallback
- anthropic-claude
- kilocode
- blackbox-free
health:
  check_interval_seconds: 300
  consecutive_failures_before_skip: 3
  recovery_check_interval_seconds: 60
  timeout_seconds: 10
heuristic_rules:
  enabled: true
  rules:
  - match:
      message_keywords:
        any_of:
        - prove
        - beweis
        - beweise
        - theorem
        - integral
        - derivative
        - equation
        - gleichung
        - induction
        - induktion
        - \boxed
        - \frac
        - \sum
        - \int
        - latex
        - step by step
        - "schritt f\xFCr schritt"
        - warum
        - why does
        - explain why
        - "erkl\xE4re warum"
        - "\u8BC1\u660E"
        - "\u5B9A\u7406"
        - "\u8A3C\u660E"
        - "\u0434\u043E\u043A\u0430\u0437\u0430\u0442\u044C"
        min_matches: 2
    name: math-reasoning
    route_to: deepseek-v4-pro
  - match:
      message_keywords:
        any_of:
        - refactor
        - debug
        - architecture
        - design pattern
        - race condition
        - memory leak
        - optimize
        - complexity
        - security vulnerability
        - code review
        - performance bottleneck
        - type error
        - deadlock
        min_matches: 2
    name: complex-code
    route_to: deepseek-v4-pro
  - match:
      message_keywords:
        any_of:
        - what is
        - was ist
        - define
        - definiere
        - translate
        - "\xFCbersetze"
        - hi
        - hello
        - hallo
        - thanks
        - danke
        - 'yes'
        - ja
        - 'no'
        - nein
        - "\u0447\u0442\u043E \u0442\u0430\u043A\u043E\u0435"
        - "\u4F60\u597D"
        - "\u3053\u3093\u306B\u3061\u306F"
        min_matches: 1
    name: simple-query
    route_to: gemini-flash-lite
  - match:
      has_tools: true
    name: tool-use
    route_to: deepseek-v4-flash
  - match:
      estimated_tokens:
        less_than: 100
    name: short-message
    route_to: deepseek-v4-flash
  - match:
      estimated_tokens:
        greater_than: 50000
    name: long-context
    route_to: deepseek-v4-flash
  - match:
      fallthrough: true
    name: general-default
    route_to: deepseek-v4-flash
llm_classifier:
  category_routing:
    AGENT: deepseek-v4-flash
    CODING: deepseek-v4-flash
    REASONING: deepseek-v4-pro
    SIMPLE: gemini-flash-lite
  classifier_provider: gemini-flash-lite
  enabled: false
  fallback_on_timeout: deepseek-v4-flash
  max_classify_tokens: 50
  prompt: 'Classify this task into exactly one category. Reply with ONLY the category
    name.

    Categories:

    - REASONING: math proofs, logic puzzles, complex analysis, debugging

    - CODING: code generation, refactoring, architecture decisions

    - SIMPLE: greetings, status checks, short lookups, translations

    - AGENT: tool use, file operations, search, API calls


    User message: {last_user_message}


    Category:

    '
  timeout_ms: 3000
metrics:
  db_path: faigate.db
  enabled: true
  log_requests: false
  log_routing_decisions: false
model_shortcuts:
  enabled: false
  shortcuts:
    anthropic-claude:
      aliases:
      - opus
      - claude-opus
      description: Anthropic Claude Opus 4.6 (premium)
      target: anthropic-claude
    anthropic-sonnet:
      aliases:
      - sonnet
      - claude-sonnet
      - claude
      description: Anthropic Claude Sonnet 4.6 (balanced)
      target: anthropic-sonnet
    anthropic-haiku:
      aliases:
      - haiku
      - claude-haiku
      description: Anthropic Claude Haiku 3.5 (fast/cheap)
      target: anthropic-haiku
    gemini-pro:
      aliases:
      - pro
      - gemini-pro-high
      description: Gemini 3.1 Pro (High) (quality)
      target: gemini-pro-high
    gemini-pro-low:
      aliases:
      - pro-low
      description: Gemini 3.1 Pro (Low) (balanced)
      target: gemini-pro-low
    blackbox-free:
      aliases:
      - blackbox
      - bb
      description: BLACKBOX budget burst route (legacy id)
      target: blackbox-free
    deepseek-v4-flash:
      aliases:
      - chat
      - ds
      description: DeepSeek Chat
      target: deepseek-v4-flash
    deepseek-v4-pro:
      aliases: []
      description: DeepSeek reasoning path
      target: deepseek-v4-pro
    flash:
      aliases:
      - gemini
      description: Gemini 3 Flash
      target: gemini-flash
    gemini-flash:
      aliases: []
      description: Gemini Flash
      target: gemini-flash
    kilocode:
      aliases:
      - kilo
      - glm5
      description: Kilo free-tier gateway path
      target: kilocode
    openai-gpt4o:
      aliases:
      - gpt4o
      - gpt-4o
      description: OpenAI GPT-4o
      target: openai-gpt4o
    openai-images:
      aliases:
      - img
      - image
      - gpt-image
      description: OpenAI image generation and editing
      target: openai-images
    r1:
      aliases:
      - reasoner
      - think
      description: DeepSeek reasoning path
      target: deepseek-v4-pro
prompt_caching:
  stabilize_prefix: false
  track_metrics: true
provider_catalog_check:
  enabled: true
  max_catalog_age_days: 30
  warn_on_model_drift: true
  warn_on_unofficial_sources: true
  warn_on_untracked: true
  warn_on_volatile_offers: true
providers:
  anthropic-haiku:
    api_key: ${ANTHROPIC_API_KEY}
    backend: anthropic-compat
    base_url: ${ANTHROPIC_BASE_URL:-https://api.anthropic.com/v1}
    capabilities:
      cost_tier: cheap
      latency_tier: fast
      streaming: true
    lane:
      benchmark_cluster: balanced-coding
      canonical_model: anthropic/haiku-3.5
      cluster: fast-workhorse
      context_strength: mid
      degrade_to:
      - deepseek/v4-flash
      - google/gemini-flash
      family: anthropic
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-24'
      name: fast
      quality_tier: mid
      reasoning_strength: mid
      review_age_days: 0
      route_type: direct
      same_model_group: anthropic/haiku-3.5
      tool_strength: medium
    max_tokens: 8096
    model: claude-haiku-3-5
    pricing:
      cache_read: 0.08
      input: 0.8
      output: 4.0
    tier: default
    timeout:
      connect_s: 10
      read_s: 45
  anthropic-sonnet:
    api_key: ${ANTHROPIC_API_KEY}
    backend: anthropic-compat
    base_url: ${ANTHROPIC_BASE_URL:-https://api.anthropic.com/v1}
    capabilities:
      cost_tier: standard
      latency_tier: balanced
      reasoning: true
      streaming: true
    lane:
      benchmark_cluster: quality-coding
      canonical_model: anthropic/sonnet-4.6
      cluster: quality-workhorse
      context_strength: high
      degrade_to:
      - anthropic/haiku-3.5
      - deepseek/v4-flash
      - google/gemini-flash
      family: anthropic
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-24'
      name: balanced
      quality_tier: high
      reasoning_strength: high
      review_age_days: 0
      route_type: direct
      same_model_group: anthropic/sonnet-4.6
      tool_strength: high
    max_tokens: 16000
    model: claude-sonnet-4-6
    pricing:
      cache_read: 0.3
      input: 3.0
      output: 15.0
    tier: mid
    timeout:
      connect_s: 10
      read_s: 90
  anthropic-claude:
    api_key: ${ANTHROPIC_API_KEY}
    backend: anthropic-compat
    base_url: ${ANTHROPIC_BASE_URL:-https://api.anthropic.com/v1}
    capabilities:
      cost_tier: premium
      latency_tier: quality
      reasoning: true
    lane:
      benchmark_cluster: quality-coding
      canonical_model: anthropic/opus-4.6
      cluster: elite-reasoning
      context_strength: high
      degrade_to:
      - anthropic/sonnet-4.6
      - openai/gpt-4o
      - deepseek/v4-pro
      family: anthropic
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-22'
      name: quality
      quality_tier: premium
      reasoning_strength: high
      review_age_days: 1
      route_type: direct
      same_model_group: anthropic/opus-4.6
      tool_strength: medium
    max_tokens: 64000
    model: claude-opus-4-6
    tier: mid
    timeout:
      connect_s: 10
      read_s: 120
  blackbox-free:
    api_key: ${BLACKBOX_API_KEY}
    backend: openai-compat
    base_url: ${BLACKBOX_BASE_URL:-https://api.blackbox.ai}
    capabilities:
      cost_tier: cheap
      latency_tier: fast
    lane:
      benchmark_cluster: budget-chat
      canonical_model: aggregator/blackbox-grok-code-fast
      cluster: budget-general
      context_strength: mid
      degrade_to:
      - aggregator/kilo-glm5-free
      - google/gemini-flash-lite
      family: blackbox
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-22'
      name: burst
      quality_tier: budget
      reasoning_strength: mid
      review_age_days: 1
      route_type: aggregator
      same_model_group: aggregator/blackbox-grok-code-fast
      tool_strength: mid
    max_tokens: 8000
    model: blackboxai/x-ai/grok-code-fast-1
    tier: fallback
    timeout:
      connect_s: 10
      read_s: 60
  deepseek-v4-flash:
    api_key: ${DEEPSEEK_API_KEY}
    backend: openai-compat
    base_url: ${DEEPSEEK_BASE_URL:-https://api.deepseek.com/v1}
    capabilities:
      cost_tier: cheap
      latency_tier: fast
      streaming: true
    lane:
      benchmark_cluster: balanced-coding
      canonical_model: deepseek/v4-flash
      cluster: balanced-workhorse
      context_strength: high
      degrade_to:
      - google/gemini-flash
      - aggregator/openrouter-auto
      family: deepseek
      freshness_hint: DeepSeek V4 (Apr 2026) — replaces V3.2 chat lane, 1M ctx
      freshness_status: fresh
      last_reviewed: '2026-04-26'
      name: workhorse
      quality_tier: mid
      reasoning_strength: mid
      review_age_days: 0
      route_type: direct
      same_model_group: deepseek/v4-flash
      tool_strength: medium
    max_tokens: 64000
    model: deepseek-v4-flash
    pricing:
      cache_read: 0.028
      input: 0.14
      output: 0.28
    tier: default
    timeout:
      connect_s: 10
      read_s: 60
  deepseek-v4-pro:
    api_key: ${DEEPSEEK_API_KEY}
    backend: openai-compat
    base_url: ${DEEPSEEK_BASE_URL:-https://api.deepseek.com/v1}
    capabilities:
      cost_tier: standard
      latency_tier: balanced
      reasoning: true
      streaming: true
    lane:
      benchmark_cluster: reasoning-coding
      canonical_model: deepseek/v4-pro
      cluster: elite-reasoning
      context_strength: high
      degrade_to:
      - deepseek/v4-flash
      - google/gemini-pro-high
      family: deepseek
      freshness_hint: DeepSeek V4 Pro (Apr 2026) — replaces V3.2 reasoner, 1M ctx, unified reasoning
      freshness_status: fresh
      last_reviewed: '2026-04-26'
      name: reasoning
      quality_tier: premium
      reasoning_strength: high
      review_age_days: 0
      route_type: direct
      same_model_group: deepseek/v4-pro
      tool_strength: high
    max_tokens: 64000
    model: deepseek-v4-pro
    pricing:
      cache_read: 0.145
      input: 1.74
      output: 3.48
    tier: reasoning
    timeout:
      connect_s: 10
      read_s: 120
  gemini-flash:
    api_key: ${GEMINI_API_KEY}
    backend: google-genai
    base_url: ${GEMINI_BASE_URL:-https://generativelanguage.googleapis.com/v1beta}
    capabilities:
      cost_tier: cheap
      latency_tier: fast
    lane:
      benchmark_cluster: fast-general
      canonical_model: google/gemini-flash
      cluster: fast-workhorse
      context_strength: mid
      degrade_to:
      - google/gemini-flash-lite
      - deepseek/v4-flash
      family: google
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-22'
      name: fast
      quality_tier: mid
      reasoning_strength: mid
      review_age_days: 1
      route_type: direct
      same_model_group: google/gemini-flash
      tool_strength: medium
    max_tokens: 8000
    model: gemini-3-flash
    tier: mid
    timeout:
      connect_s: 10
      read_s: 60
  gemini-flash-lite:
    api_key: ${GEMINI_API_KEY}
    backend: google-genai
    base_url: ${GEMINI_BASE_URL:-https://generativelanguage.googleapis.com/v1beta}
    capabilities:
      cost_tier: cheap
      latency_tier: fast
    lane:
      benchmark_cluster: budget-chat
      canonical_model: google/gemini-flash-lite
      cluster: budget-general
      context_strength: mid
      degrade_to:
      - aggregator/kilo-glm5-free
      - aggregator/blackbox-grok-code-fast
      family: google
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-24'
      name: cheap
      quality_tier: budget
      reasoning_strength: low
      review_age_days: 0
      route_type: direct
      same_model_group: google/gemini-flash-lite
      tool_strength: low
    max_tokens: 8000
    model: gemini-3-flash-lite
    pricing:
      cache_read: 0.02
      input: 0.075
      output: 0.3
    tier: cheap
    timeout:
      connect_s: 10
      read_s: 30
  gemini-pro-high:
    api_key: ${GEMINI_API_KEY}
    backend: google-genai
    base_url: ${GEMINI_BASE_URL:-https://generativelanguage.googleapis.com/v1beta}
    capabilities:
      cost_tier: premium
      latency_tier: balanced
      reasoning: true
    lane:
      benchmark_cluster: quality-coding
      canonical_model: google/gemini-pro-high
      cluster: quality-workhorse
      context_strength: high
      degrade_to:
      - google/gemini-pro-low
      - google/gemini-flash
      - deepseek/v4-pro
      family: google
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-29'
      name: pro-high
      quality_tier: high
      reasoning_strength: high
      review_age_days: 0
      route_type: direct
      same_model_group: google/gemini-pro-high
      tool_strength: high
    max_tokens: 65536
    model: gemini-3.1-pro
    pricing:
      cache_read: 0.31
      input: 1.25
      output: 10.0
    tier: mid
    timeout:
      connect_s: 10
      read_s: 120
  gemini-pro-low:
    api_key: ${GEMINI_API_KEY}
    backend: google-genai
    base_url: ${GEMINI_BASE_URL:-https://generativelanguage.googleapis.com/v1beta}
    capabilities:
      cost_tier: premium
      latency_tier: balanced
      reasoning: true
    lane:
      benchmark_cluster: balanced-coding
      canonical_model: google/gemini-pro-low
      cluster: balanced-workhorse
      context_strength: high
      degrade_to:
      - google/gemini-flash
      - deepseek/v4-flash
      family: google
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-29'
      name: pro-low
      quality_tier: mid
      reasoning_strength: mid
      review_age_days: 0
      route_type: direct
      same_model_group: google/gemini-pro-low
      tool_strength: medium
    max_tokens: 65536
    model: gemini-3.1-pro
    pricing:
      cache_read: 0.20
      input: 0.80
      output: 4.0
    tier: mid
    timeout:
      connect_s: 10
      read_s: 90
  kilocode:
    api_key: ${KILOCODE_API_KEY}
    backend: openai-compat
    base_url: ${KILOCODE_BASE_URL:-https://api.kilo.ai/api/gateway}
    capabilities:
      cost_tier: free
      latency_tier: balanced
    lane:
      benchmark_cluster: free-coding
      canonical_model: aggregator/kilo-glm5-free
      cluster: budget-general
      context_strength: mid
      degrade_to:
      - aggregator/blackbox-grok-code-fast
      - google/gemini-flash-lite
      family: kilo
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-22'
      name: free
      quality_tier: free
      reasoning_strength: mid
      review_age_days: 1
      route_type: aggregator
      same_model_group: aggregator/kilo-glm5-free
      tool_strength: low
    max_tokens: 8000
    model: z-ai/glm-5:free
    tier: fallback
    timeout:
      connect_s: 10
      read_s: 60
  kilo-auto/frontier:
    api_key: ${KILOCODE_API_KEY}
    backend: openai-compat
    base_url: ${KILOCODE_BASE_URL:-https://api.kilo.ai/api/gateway}
    capabilities:
      cost_tier: premium
      latency_tier: balanced
    max_tokens: 32000
    model: kilo-auto/frontier
    tier: premium
    timeout:
      connect_s: 10
      read_s: 120
  kilo-auto/balanced:
    api_key: ${KILOCODE_API_KEY}
    backend: openai-compat
    base_url: ${KILOCODE_BASE_URL:-https://api.kilo.ai/api/gateway}
    capabilities:
      cost_tier: mid
      latency_tier: balanced
    max_tokens: 16000
    model: kilo-auto/balanced
    tier: mid
    timeout:
      connect_s: 10
      read_s: 90
  kilo-auto/free:
    api_key: ${KILOCODE_API_KEY}
    backend: openai-compat
    base_url: ${KILOCODE_BASE_URL:-https://api.kilo.ai/api/gateway}
    capabilities:
      cost_tier: free
      latency_tier: balanced
    max_tokens: 8000
    model: kilo-auto/free
    tier: fallback
    timeout:
      connect_s: 10
      read_s: 60
  kilo-auto/small:
    api_key: ${KILOCODE_API_KEY}
    backend: openai-compat
    base_url: ${KILOCODE_BASE_URL:-https://api.kilo.ai/api/gateway}
    capabilities:
      cost_tier: cheap
      latency_tier: fast
    max_tokens: 8000
    model: kilo-auto/small
    tier: cheap
    timeout:
      connect_s: 10
      read_s: 60
  openai-gpt4o:
    api_key: ${OPENAI_API_KEY}
    backend: openai-compat
    base_url: ${OPENAI_BASE_URL:-https://api.openai.com/v1}
    capabilities:
      cost_tier: premium
      latency_tier: balanced
    lane:
      benchmark_cluster: quality-coding
      canonical_model: openai/gpt-4o
      cluster: quality-workhorse
      context_strength: high
      degrade_to:
      - openai/gpt-4o-mini
      - google/gemini-pro-high
      family: openai
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-22'
      name: balanced
      quality_tier: high
      reasoning_strength: high
      review_age_days: 1
      route_type: direct
      same_model_group: openai/gpt-4o
      tool_strength: high
    max_tokens: 8192
    model: gpt-4o
    tier: mid
    timeout:
      connect_s: 10
      read_s: 60
  openai-images:
    api_key: ${OPENAI_API_KEY}
    backend: openai-compat
    base_url: ${OPENAI_BASE_URL:-https://api.openai.com/v1}
    capabilities:
      cost_tier: premium
      image_editing: true
    contract: image-provider
    image:
      max_outputs: 4
      max_side_px: 2048
      policy_tags:
      - quality
      - editing
      - batch
      supported_sizes:
      - 1024x1024
      - 1536x1024
      - 1024x1536
    model: gpt-image-1
    tier: specialty
  openrouter-fallback:
    api_key: ${OPENROUTER_API_KEY}
    backend: openai-compat
    base_url: ${OPENROUTER_BASE_URL:-https://openrouter.ai/api/v1}
    capabilities:
      cost_tier: marketplace
      latency_tier: balanced
      streaming: true
    lane:
      benchmark_cluster: marketplace-general
      canonical_model: aggregator/openrouter-auto
      cluster: aggregator-fallback
      context_strength: variable
      degrade_to:
      - aggregator/kilo-glm5-free
      - aggregator/blackbox-grok-code-fast
      family: openrouter
      freshness_hint: benchmark and cost assumptions were reviewed recently
      freshness_status: fresh
      last_reviewed: '2026-03-22'
      name: router
      quality_tier: variable
      reasoning_strength: variable
      review_age_days: 1
      route_type: aggregator
      same_model_group: aggregator/openrouter-auto
      tool_strength: variable
    max_tokens: 8000
    model: openrouter/auto
    tier: fallback
    timeout:
      connect_s: 10
      read_s: 90

  # ── Free Providers (OmniRoute-inspired zero-cost fallback stack) ────────────
  pollinations:
    api_key: ""
    backend: openai-compat
    base_url: https://text.pollinations.ai/openai
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
    lane:
      benchmark_cluster: budget-chat
      canonical_model: aggregator/pollinations-openai
      cluster: budget-general
      context_strength: mid
      degrade_to: []
      family: pollinations
      freshness_hint: free no-key provider — polling models endpoint for availability
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: variable
      review_age_days: 0
      route_type: aggregator
      same_model_group: aggregator/pollinations-openai
      tool_strength: low
    max_tokens: 8000
    model: openai
    tier: free
    timeout:
      connect_s: 10
      read_s: 60
    transport:
      auth_mode: none
      profile: free-generative
      requires_api_key: false
  groq:
    api_key: ${GROQ_API_KEY}
    backend: openai-compat
    base_url: https://api.groq.com/openai/v1
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: true
    lane:
      benchmark_cluster: budget-chat
      canonical_model: groq/llama-3.3-70b
      cluster: budget-general
      context_strength: mid
      degrade_to: []
      family: groq
      freshness_hint: free tier — 30 RPM, 14,400 RPD, 6K TPM
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: mid
      review_age_days: 0
      route_type: direct
      same_model_group: groq/llama-3.3-70b
      tool_strength: low
    max_tokens: 8000
    model: llama-3.3-70b-versatile
    tier: free
    timeout:
      connect_s: 10
      read_s: 30
  cerebras:
    api_key: ${CEREBRAS_API_KEY}
    backend: openai-compat
    base_url: https://api.cerebras.ai/v1
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: true
      reasoning: true
    lane:
      benchmark_cluster: budget-chat
      canonical_model: cerebras/qwen3-235b
      cluster: budget-general
      context_strength: mid
      degrade_to: []
      family: cerebras
      freshness_hint: free tier — 30 RPM, 1M TPD
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: high
      review_age_days: 0
      route_type: direct
      same_model_group: cerebras/qwen3-235b
      tool_strength: medium
    max_tokens: 8000
    model: qwen-3-235b-a22b-instruct-2507
    tier: free
    timeout:
      connect_s: 10
      read_s: 60
  longcat:
    api_key: ${LONGCAT_API_KEY}
    backend: openai-compat
    base_url: https://api.longcat.ai/v1
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
    lane:
      benchmark_cluster: budget-chat
      canonical_model: longcat/flash-lite
      cluster: budget-general
      context_strength: mid
      degrade_to: []
      family: longcat
      freshness_hint: free tier — 50M tokens/day
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: low
      review_age_days: 0
      route_type: direct
      same_model_group: longcat/flash-lite
      tool_strength: low
    max_tokens: 8000
    model: longcat-flash-lite
    tier: free
    timeout:
      connect_s: 10
      read_s: 60
  nvidia-nim:
    api_key: ${NVIDIA_API_KEY}
    backend: openai-compat
    base_url: https://integrate.api.nvidia.com/v1
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: true
      reasoning: true
    lane:
      benchmark_cluster: budget-chat
      canonical_model: nvidia/nemotron
      cluster: budget-general
      context_strength: mid
      degrade_to: []
      family: nvidia
      freshness_hint: free tier — 129 models, 40 RPM
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: high
      review_age_days: 0
      route_type: direct
      same_model_group: nvidia/nemotron
      tool_strength: medium
    max_tokens: 8000
    model: nvidia/nemotron-super-49b-v1
    tier: free
    timeout:
      connect_s: 10
      read_s: 60
  kiro:
    api_key: ${KIRO_API_KEY:-}
    backend: openai-compat
    base_url: ${KIRO_BASE_URL:-https://api.kiro.dev/v1}
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
    lane:
      benchmark_cluster: budget-chat
      canonical_model: aggregator/kiro-claude-sonnet
      cluster: budget-general
      context_strength: high
      degrade_to: []
      family: kiro
      freshness_hint: free via AWS Builder ID OAuth — Claude Sonnet 4.5 unlimited
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: high
      review_age_days: 0
      route_type: aggregator
      same_model_group: aggregator/kiro-claude-sonnet
      tool_strength: high
    max_tokens: 8000
    model: claude-sonnet-4-6
    tier: free
    timeout:
      connect_s: 10
      read_s: 90
    transport:
      auth_mode: none
      profile: free-generative
      requires_api_key: false
  qoder:
    api_key: ${QODER_API_KEY:-}
    backend: openai-compat
    base_url: ${QODER_BASE_URL:-https://api.qoder.ai/v1}
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
      reasoning: true
    lane:
      benchmark_cluster: free-coding
      canonical_model: aggregator/qoder-kimi-k2
      cluster: budget-general
      context_strength: high
      degrade_to: []
      family: qoder
      freshness_hint: free via Google OAuth — kimi-k2-thinking unlimited
      freshness_status: fresh
      last_reviewed: '2026-05-04'
      name: free
      quality_tier: free
      reasoning_strength: high
      review_age_days: 0
      route_type: aggregator
      same_model_group: aggregator/qoder-kimi-k2
      tool_strength: medium
    max_tokens: 8000
    model: kimi-k2-thinking
    tier: free
    timeout:
      connect_s: 10
      read_s: 90
    transport:
      auth_mode: none
      profile: free-generative
      requires_api_key: false

  # ── Additional API-key providers ──────────────────────────────────────────

  # Cohere – Command A/R series, OpenAI-compat compatibility endpoint
  cohere:
    api_key: ${COHERE_API_KEY}
    backend: openai-compat
    base_url: ${COHERE_BASE_URL:-https://api.cohere.com/compatibility/v1}
    model: command-a-03-2025
    tier: default
    timeout:
      connect_s: 10
      read_s: 60

  # Mistral AI – Mistral Large, Codestral, etc.
  mistral:
    api_key: ${MISTRAL_API_KEY}
    backend: openai-compat
    base_url: ${MISTRAL_BASE_URL:-https://api.mistral.ai/v1}
    model: mistral-large-latest
    tier: default
    timeout:
      connect_s: 10
      read_s: 90

  # ── Local runtimes (uncomment and configure) ──────────────────────────────
  # ollama:
  #   api_key: ""
  #   backend: openai-compat
  #   base_url: http://127.0.0.1:11434/v1
  #   auth_optional: true
  #   capabilities:
  #     cost_tier: local
  #     latency_tier: local
  #   max_tokens: 8192
  #   model: ollama/llama3.3
  #   tier: local
  #   timeout:
  #     connect_s: 10
  #     read_s: 120

  # vllm:
  #   api_key: ""
  #   backend: openai-compat
  #   base_url: http://127.0.0.1:8000/v1
  #   auth_optional: true
  #   capabilities:
  #     cost_tier: local
  #     latency_tier: local
  #   max_tokens: 8192
  #   model: vllm/your-model-id
  #   tier: local
  #   timeout:
  #     connect_s: 10
  #     read_s: 120

  # lmstudio:
  #   api_key: ""
  #   backend: openai-compat
  #   base_url: http://localhost:1234/v1
  #   auth_optional: true
  #   capabilities:
  #     cost_tier: local
  #     latency_tier: local
  #   max_tokens: 8192
  #   model: lmstudio/minimax-m2.1-gs32
  #   tier: local
  #   timeout:
  #     connect_s: 10
  #     read_s: 120

  # litellm:
  #   api_key: ""
  #   backend: openai-compat
  #   base_url: http://localhost:4000/v1
  #   auth_optional: true
  #   capabilities:
  #     cost_tier: local
  #     latency_tier: local
  #   max_tokens: 8192
  #   model: litellm/your-model-id
  #   tier: local
  #   timeout:
  #     connect_s: 10
  #     read_s: 120

  # ── Generic providers (use any model) ─────────────────────────────────────
  # openai:
  #   api_key: ${OPENAI_API_KEY}
  #   backend: openai-compat
  #   base_url: ${OPENAI_BASE_URL:-https://api.openai.com/v1}
  #   capabilities:
  #     cost_tier: standard
  #     latency_tier: balanced
  #   max_tokens: 8192
  #   model: gpt-4o  # any OpenAI model ID
  #   tier: default
  #   timeout:
  #     connect_s: 10
  #     read_s: 60

  # anthropic:
  #   api_key: ${ANTHROPIC_API_KEY}
  #   backend: anthropic-compat
  #   base_url: ${ANTHROPIC_BASE_URL:-https://api.anthropic.com/v1}
  #   capabilities:
  #     cost_tier: standard
  #     latency_tier: balanced
  #   max_tokens: 16000
  #   model: claude-sonnet-4-6  # any Anthropic model ID
  #   tier: default
  #   timeout:
  #     connect_s: 10
  #     read_s: 90

  # google:
  #   api_key: ${GEMINI_API_KEY}
  #   backend: google-genai
  #   base_url: ${GEMINI_BASE_URL:-https://generativelanguage.googleapis.com/v1beta}
  #   capabilities:
  #     cost_tier: standard
  #     latency_tier: balanced
  #   max_tokens: 65536
  #   model: gemini-3.1-pro  # any Gemini model ID
  #   tier: default
  #   timeout:
  #     connect_s: 10
  #     read_s: 90

  # ── OAuth‑managed providers ────────────────────────────────────────────────
  #
  # qwen-portal (Qwen OAuth free tier via qwen-code CLI)
  # Prerequisites: faigate-auth qwen-portal  (browser flow, no node required)
  # Token read from ~/.qwen/oauth_creds.json; auto-refresh on expiry.
  qwen-portal:
    backend: oauth
    oauth:
      helper: "faigate-auth qwen-portal"
      client_id: "f0304373b74a44d2b584a3fb70ca9e56"
      token_endpoint: "https://chat.qwen.ai/api/v1/oauth2/token"
      refresh_endpoint: "https://chat.qwen.ai/api/v1/oauth2/token"
      scope: "openid profile email model.completion"
    underlying_backend: openai-compat
    base_url: "https://portal.qwen.ai/v1"
    model: coder-model
    tier: default
    # portal.qwen.ai requires a system message in every request
    require_system_message: "You are a helpful assistant."
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
    # Required headers for portal.qwen.ai (same as qwen-code CLI sends)
    extra_headers:
      user-agent: "QwenCode/0.14.0 (darwin; arm64)"
      x-dashscope-authtype: "qwen-oauth"
      x-dashscope-cachecontrol: "enable"
      x-dashscope-useragent: "QwenCode/0.14.0 (darwin; arm64)"
      x-stainless-arch: "arm64"
      x-stainless-lang: "js"
      x-stainless-os: "MacOS"
      x-stainless-package-version: "5.11.0"
      x-stainless-runtime: "node"
      x-stainless-runtime-version: "v25.9.0"
    timeout:
      connect_s: 10
      read_s: 90

  claude-code:
    # Token from macOS Keychain (Claude Code-credentials). Run: faigate-auth claude-code
    # Inference via api.anthropic.com/v1/messages with Bearer OAuth token.
    # Requires anthropic-beta: oauth-2025-04-20 header (injected via extra_headers).
    # Refresh tokens are single-use; faigate rewrites keychain after every refresh.
    backend: oauth
    oauth:
      helper: "faigate-auth claude-code"
      client_id: "9d1c250a-e61b-44d9-88ed-5944d1962f5e"
      token_endpoint: "https://platform.claude.com/v1/oauth/token"
      refresh_endpoint: "https://platform.claude.com/v1/oauth/token"
      scope: "user:inference user:profile"
    underlying_backend: anthropic-compat
    base_url: "https://api.anthropic.com/v1"
    model: claude-sonnet-4-6
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
    extra_headers:
      anthropic-version: "2023-06-01"
      anthropic-beta: "oauth-2025-04-20"
      anthropic-dangerous-direct-browser-access: "true"
    timeout:
      connect_s: 10
      read_s: 120

  openai-codex:
    # Token from ~/.codex/auth.json (auth_mode=chatgpt). Run: faigate-auth openai-codex
    # Inference endpoint: chatgpt.com/backend-api/codex/responses (NOT api.openai.com)
    # Refresh tokens are single-use; faigate rewrites auth.json after every refresh.
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.4
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: balanced
      reasoning: true
      streaming: true
    lane:
      canonical_model: openai-codex/gpt-5.4
      cluster: elite-reasoning
      benchmark_cluster: quality-coding
      quality_tier: premium
      reasoning_strength: high
      context_strength: high
      tool_strength: medium
      route_type: direct
      name: codex
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 90
  openai-codex-mini:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.1-codex-mini
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: fast
      reasoning: true
      streaming: true
    lane:
      canonical_model: openai-codex/gpt-5.1-codex-mini
      cluster: fast-workhorse
      benchmark_cluster: fast-coding
      quality_tier: mid
      reasoning_strength: mid
      context_strength: high
      tool_strength: medium
      route_type: direct
      name: codex-mini
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 60
  openai-codex-spark:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.3-codex-spark
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: true
    lane:
      canonical_model: openai-codex/gpt-5.3-codex-spark
      cluster: fast-workhorse
      benchmark_cluster: fast-coding
      quality_tier: mid
      route_type: direct
      name: codex-spark
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 60
  openai-codex-high:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.3-codex-high
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: balanced
      reasoning: true
      streaming: true
    lane:
      canonical_model: openai-codex/gpt-5.3-codex-high
      cluster: elite-reasoning
      benchmark_cluster: quality-coding
      quality_tier: premium
      reasoning_strength: high
      context_strength: high
      tool_strength: medium
      route_type: direct
      name: codex-high
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 90
  openai-codex-xhigh:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.3-codex-xhigh
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: quality
      reasoning: true
      streaming: true
    lane:
      canonical_model: openai-codex/gpt-5.3-codex-xhigh
      cluster: elite-reasoning
      benchmark_cluster: quality-coding
      quality_tier: premium
      reasoning_strength: high
      context_strength: high
      tool_strength: medium
      route_type: direct
      name: codex-xhigh
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 120
  openai-codex-low:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.3-codex-low
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: true
    lane:
      canonical_model: openai-codex/gpt-5.3-codex-low
      cluster: fast-workhorse
      benchmark_cluster: fast-coding
      quality_tier: mid
      reasoning_strength: low
      context_strength: high
      tool_strength: medium
      route_type: direct
      name: codex-low
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 60
  # GPT-5.4 with explicit reasoning_effort (injected via extra_body)
  openai-codex-5.4-xhigh:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.4
    extra_body:
      reasoning_effort: extra_high
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: quality
      reasoning: true
      streaming: true
    lane:
      family: openai
      canonical_model: openai-codex/gpt-5.4
      cluster: elite-reasoning
      benchmark_cluster: quality-coding
      quality_tier: premium
      reasoning_strength: high
      route_type: direct
      name: codex-5.4-xhigh
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 120
  openai-codex-5.4-high:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.4
    extra_body:
      reasoning_effort: high
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: balanced
      reasoning: true
      streaming: true
    lane:
      family: openai
      canonical_model: openai-codex/gpt-5.4
      cluster: elite-reasoning
      benchmark_cluster: quality-coding
      quality_tier: premium
      reasoning_strength: high
      route_type: direct
      name: codex-5.4-high
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 90
  openai-codex-5.4-medium:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.4
    extra_body:
      reasoning_effort: medium
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: true
    lane:
      family: openai
      canonical_model: openai-codex/gpt-5.4
      cluster: quality-workhorse
      benchmark_cluster: quality-coding
      quality_tier: high
      reasoning_strength: mid
      route_type: direct
      name: codex-5.4-medium
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 60
  openai-codex-5.4-low:
    backend: oauth
    oauth:
      helper: "faigate-auth openai-codex"
      client_id: "app_EMoamEEZ73f0CkXaXp7hrann"
      token_endpoint: "https://auth.openai.com/oauth/token"
      refresh_endpoint: "https://auth.openai.com/oauth/token"
      scope: "openid profile email offline_access"
    underlying_backend: openai-compat
    base_url: "https://chatgpt.com/backend-api/codex/responses"
    model: gpt-5.4
    extra_body:
      reasoning_effort: low
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: true
    lane:
      family: openai
      canonical_model: openai-codex/gpt-5.4
      cluster: fast-workhorse
      benchmark_cluster: fast-coding
      quality_tier: mid
      reasoning_strength: low
      route_type: direct
      name: codex-5.4-low
    pricing:
      input: 0.0
      output: 0.0
    timeout:
      connect_s: 10
      read_s: 45

  # Antigravity / Gemini CLI (Google OAuth → cloudcode-pa.googleapis.com free tier)
  # Both tools share ~/.gemini/oauth_creds.json. Token is refreshed automatically by
  # the gemini CLI binary; faigate reads it via faigate-auth google-antigravity.
  #
  # cloudcode-pa API: custom Gemini wrapper with {model, project, request: <gemini_body>}
  #   Backend: google-codeassist  (Bearer auth, wraps Gemini format)
  #   Endpoint: https://cloudcode-pa.googleapis.com/v1internal
  #   google_project: your Google project ID (from gemini-cli token exchange)
  #
  # Prerequisites: gemini CLI must be authenticated (gemini auth login)
  #   Token file: ~/.gemini/oauth_creds.json
  #
  antigravity:
    backend: oauth
    oauth:
      helper: "faigate-auth google-antigravity"
      client_id: "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
      token_endpoint: "https://oauth2.googleapis.com/token"
      refresh_endpoint: "https://oauth2.googleapis.com/token"
    underlying_backend: google-codeassist
    base_url: "https://cloudcode-pa.googleapis.com/v1internal"
    google_project: "yarling-bloom-d80f5"
    model: gemini-2.5-pro
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: balanced
      streaming: false
    timeout:
      connect_s: 10
      read_s: 90

  gemini-cli:
    backend: oauth
    oauth:
      helper: "faigate-auth google-antigravity"
      client_id: "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
      token_endpoint: "https://oauth2.googleapis.com/token"
      refresh_endpoint: "https://oauth2.googleapis.com/token"
    underlying_backend: google-codeassist
    base_url: "https://cloudcode-pa.googleapis.com/v1internal"
    google_project: "yarling-bloom-d80f5"
    model: gemini-2.5-flash-lite
    tier: default
    capabilities:
      cost_tier: free
      latency_tier: fast
      streaming: false
    timeout:
      connect_s: 10
      read_s: 60

  # ---------------------------------------------------------------------------
  # Additional providers (LLMAIRouter parity) – uncomment to activate
  # ---------------------------------------------------------------------------

  # DeepSeek – NOTE: deepseek-v4-flash and deepseek-v4-pro are already configured
  # above as active providers. This entry is a convenience alias using the
  # registry key; the active configs above take precedence.
  # deepseek:
  #   api_key: ${DEEPSEEK_API_KEY}
  #   backend: openai-compat
  #   base_url: ${DEEPSEEK_BASE_URL:-https://api.deepseek.com/v1}
  #   model: deepseek-v4-pro
  #   tier: reasoning
  #   timeout:
  #     connect_s: 10
  #     read_s: 120

  # Together AI – serverless inference, Llama / Mixtral / DeepSeek / Qwen
  # together:
  #   api_key: ${TOGETHER_API_KEY}
  #   backend: openai-compat
  #   base_url: ${TOGETHER_BASE_URL:-https://api.together.xyz/v1}
  #   model: meta-llama/Llama-3.3-70B-Instruct-Turbo
  #   tier: cheap
  #   timeout:
  #     connect_s: 10
  #     read_s: 60

  # Fireworks AI – fast serverless inference, DeepSeek / Llama / Qwen
  # fireworks:
  #   api_key: ${FIREWORKS_API_KEY}
  #   backend: openai-compat
  #   base_url: ${FIREWORKS_BASE_URL:-https://api.fireworks.ai/inference/v1}
  #   model: accounts/fireworks/models/deepseek-r1
  #   tier: cheap
  #   timeout:
  #     connect_s: 10
  #     read_s: 90

  # Cohere – Command A/R series, OpenAI-compat compatibility endpoint
  # cohere:
  #   api_key: ${COHERE_API_KEY}
  #   backend: openai-compat
  #   base_url: ${COHERE_BASE_URL:-https://api.cohere.com/compatibility/v1}
  #   model: command-a-03-2025
  #   tier: default
  #   timeout:
  #     connect_s: 10
  #     read_s: 60

  # Nebius AI Studio – DeepSeek / Llama / Qwen on European infra
  # nebius:
  #   api_key: ${NEBIUS_API_KEY}
  #   backend: openai-compat
  #   base_url: ${NEBIUS_BASE_URL:-https://api.studio.nebius.ai/v1}
  #   model: deepseek-ai/DeepSeek-R1
  #   tier: cheap
  #   timeout:
  #     connect_s: 10
  #     read_s: 90

  # SiliconFlow – low-cost inference (CN), DeepSeek / Qwen / GLM
  # siliconflow:
  #   api_key: ${SILICONFLOW_API_KEY}
  #   backend: openai-compat
  #   base_url: ${SILICONFLOW_BASE_URL:-https://api.siliconflow.cn/v1}
  #   model: deepseek-ai/DeepSeek-R1
  #   tier: cheap
  #   timeout:
  #     connect_s: 10
  #     read_s: 90

  # Hyperbolic – GPU cloud inference, DeepSeek / Llama / Qwen
  # hyperbolic:
  #   api_key: ${HYPERBOLIC_API_KEY}
  #   backend: openai-compat
  #   base_url: ${HYPERBOLIC_BASE_URL:-https://api.hyperbolic.xyz/v1}
  #   model: deepseek-ai/DeepSeek-R1
  #   tier: cheap
  #   timeout:
  #     connect_s: 10
  #     read_s: 90

  # Perplexity – online/search-augmented models
  # perplexity:
  #   api_key: ${PERPLEXITY_API_KEY}
  #   backend: openai-compat
  #   base_url: ${PERPLEXITY_BASE_URL:-https://api.perplexity.ai}
  #   model: sonar-pro
  #   tier: default
  #   timeout:
  #     connect_s: 10
  #     read_s: 60

  # NVIDIA NIM – optimized inference on NVIDIA infra
  # nvidia-nim:
  #   api_key: ${NVIDIA_API_KEY}
  #   backend: openai-compat
  #   base_url: ${NVIDIA_NIM_BASE_URL:-https://integrate.api.nvidia.com/v1}
  #   model: deepseek-ai/deepseek-r1
  #   tier: default
  #   timeout:
  #     connect_s: 10
  #     read_s: 90
request_hooks:
  enabled: true
  hooks:
  - prefer-provider-header
  - locality-header
  - profile-override-header
  - mode-override-header
  on_error: continue
api_surfaces:
  anthropic_messages: true
anthropic_bridge:
  enabled: true
  allow_claude_code_hints: true
  model_aliases:
    claude-code: auto
    claude-code-fast: eco
    claude-code-premium: premium
    claude-sonnet-4-6: auto
    claude-sonnet-4-6-20251001: auto
    claude-sonnet-4-6[1m]: auto
    claude-opus-4-6: premium
    claude-opus-4-6-20251001: premium
    claude-opus-4-6[1m]: premium
    claude-haiku-4-5: eco
    claude-haiku-4-5-20251001: eco
    # Claude Desktop model aliases
    claude-3-5-sonnet-20241022: auto
    claude-3-5-sonnet: auto
    claude-3-opus-20240229: premium
    claude-3-opus: premium
    claude-3-haiku-20240307: eco
    claude-3-haiku: eco
routing_modes:
  default: auto
  enabled: true
  modes:
    auto:
      aliases: []
      best_for: General use
      description: Balanced (default)
      savings: Balanced
      select:
        prefer_tiers:
        - default
        - reasoning
        - mid
    coding-auto:
      aliases:
      - code-auto
      - dev-auto
      best_for: Daily coding with strong cost control
      description: Cheapest capable coding route
      savings: High savings
      select:
        capability_values:
          cost_tier:
          - free
          - cheap
          - standard
        prefer_tiers:
        - default
        - cheap
        - fallback
        - reasoning
        - mid
        routing_mode: eco
    coding-fast:
      aliases:
      - code-fast
      - dev-fast
      best_for: Fast iterative coding loops
      description: Speed-first coding workhorse
      savings: High savings
      select:
        capability_values:
          cost_tier:
          - free
          - cheap
          - standard
        prefer_tiers:
        - cheap
        - default
        - fallback
        prefer_providers:
        - groq
        - cerebras
        - gemini-flash-lite
        - gemini-flash
        - anthropic-haiku
        - deepseek-v4-flash
        routing_mode: eco
    coding-premium:
      aliases:
      - code-premium
      - dev-premium
      best_for: Architecture, review, and high-risk coding changes
      description: Strongest coding lanes with cost still visible
      savings: Lowest savings
      select:
        prefer_tiers:
        - reasoning
        - default
        - mid
        - high
        routing_mode: premium
    eco:
      aliases: []
      best_for: Maximum savings
      description: Cheapest possible
      savings: Aggressive savings
      select:
        capability_values:
          cost_tier:
          - free
          - cheap
          - standard
        prefer_providers:
        - pollinations
        - groq
        - cerebras
        - anthropic-haiku
        - gemini-flash-lite
        - gemini-flash
        - deepseek-v4-flash
        prefer_tiers:
        - cheap
        - fallback
        - default
    free:
      aliases: []
      best_for: Zero-cost routing
      description: Free tier only
      savings: 100%
      select:
        allow_providers: &id001
        - kilocode
        - blackbox-free
        capability_values:
          cost_tier:
          - free
        prefer_providers: *id001
    premium:
      aliases: []
      best_for: Mission-critical work
      description: Best quality
      savings: Lowest savings
      select:
        prefer_providers:
        - anthropic-sonnet
        - openai-gpt4o
        - gemini-pro-high
        - anthropic-claude
        - deepseek-v4-pro
        prefer_tiers:
        - high
        - mid
        - reasoning
routing_policies:
  enabled: false
  rules:
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - n8n
      - client_profile:
        - n8n
    name: n8n-cheap-default
    select:
      prefer_tiers:
      - cheap
      - default
      - mid
  - match:
      any:
      - header_contains:
          x-faigate-client:
          - opencode
      - client_profile:
        - opencode
    name: opencode-general
    select:
      prefer_tiers:
      - default
      - mid
      - reasoning
security:
  cache_control: no-store
  max_header_value_chars: 160
  max_json_body_bytes: 1048576
  max_upload_bytes: 10485760
  response_headers: true
server:
  host: 127.0.0.1
  log_level: info
  port: 8092
static_rules:
  enabled: true
  rules:
  - match:
      any:
      - system_prompt_contains:
        - heartbeat
        - health check
        - status check
        - are you there
      - model_requested:
        - heartbeat
        - cheap
        - flash-lite
    name: heartbeat
    route_to: gemini-flash-lite
  - match:
      model_requested:
      - reasoner
      - r1
      - think
      - deepseek-v4-pro
    name: explicit-reasoner
    route_to: deepseek-v4-pro
  - match:
      model_requested:
      - flash
      - gemini
      - vision
    name: explicit-flash
    route_to: gemini-flash
  - match:
      model_requested:
      - chat
      - ds
      - default
      - deepseek-v4-flash
    name: explicit-chat
    route_to: deepseek-v4-flash
  - match:
      model_requested:
      - codex-mini
      - openai-codex-mini
      - gpt-5.1-codex-mini
      - codex-spark
      - openai-codex-spark
      - gpt-5.3-codex-spark
    name: explicit-codex-mini
    route_to: openai-codex-mini
  - match:
      model_requested:
      - codex-xhigh
      - openai-codex-xhigh
      - gpt-5.3-codex-xhigh
    name: explicit-codex-xhigh
    route_to: openai-codex-xhigh
  - match:
      model_requested:
      - codex-high
      - openai-codex-high
      - gpt-5.3-codex-high
    name: explicit-codex-high
    route_to: openai-codex-high
  - match:
      model_requested:
      - codex-low
      - openai-codex-low
      - gpt-5.3-codex-low
    name: explicit-codex-low
    route_to: openai-codex-low
  - match:
      model_requested:
      - codex-5.4-xhigh
      - openai-codex-5.4-xhigh
      - gpt-5.4-xhigh
    name: explicit-codex-5.4-xhigh
    route_to: openai-codex-5.4-xhigh
  - match:
      model_requested:
      - codex-5.4-high
      - openai-codex-5.4-high
      - gpt-5.4-high
    name: explicit-codex-5.4-high
    route_to: openai-codex-5.4-high
  - match:
      model_requested:
      - codex-5.4-medium
      - openai-codex-5.4-medium
      - gpt-5.4-medium
    name: explicit-codex-5.4-medium
    route_to: openai-codex-5.4-medium
  - match:
      model_requested:
      - codex-5.4-low
      - openai-codex-5.4-low
      - gpt-5.4-low
    name: explicit-codex-5.4-low
    route_to: openai-codex-5.4-low
  - match:
      model_requested:
      - codex
      - openai-codex
      - gpt-5.4
      - gpt-5.3-codex
    name: explicit-codex
    route_to: openai-codex
  - match:
      any:
      - system_prompt_contains:
        - sub-agent
        - subagent
        - parallel task
        - delegated task
      - header_contains:
          x-openclaw-source:
          - subagent
          - sub-agent
    name: subagent
    route_to: deepseek-v4-flash
update_check:
  api_base: https://api.github.com
  check_interval_seconds: 21600
  enabled: true
  release_channel: stable
  repository: fusionAIze/faigate
  timeout_seconds: 5