config.yaml

# Quantum Code Model Definitions
# Runtime defaults are in Settings class (quantum_code/settings.py)
# API keys should be in .env file

version: "1.0"

# =============================================================================
# Model Definitions
# =============================================================================
models:
  # ============================================
  # OpenAI Models
  # ============================================

  gpt-5-nano:
    litellm_model: openai/gpt-5-nano
    aliases:
      - nano
    constraints:
      temperature: 1.0  # gpt-5 family only supports 1.0
    notes: "Fast, cost-effective"

  gpt-5-mini:
    litellm_model: openai/gpt-5-mini
    provider_web_search: true  # Native web search support (per OpenAI docs)
    aliases:
      - mini
    constraints:
      temperature: 1.0  # gpt-5 family only supports 1.0
    notes: "Fast, cost-effective with web search"
    # disabled: true  # Uncomment this line to disable this model (it won't appear in /multi:models)

  gpt-5.2:
    litellm_model: openai/gpt-5.2
    provider_web_search: true  # Native web search support (per OpenAI docs)
    aliases:
      - gpt-5
      - gpt
    constraints:
      temperature: 1.0
    notes: "Strong general-purpose model with web search, temperature=1.0 only"

  gpt-5.1:
    litellm_model: openai/gpt-5.1
    provider_web_search: true  # Native web search support (per OpenAI docs)
    constraints:
      temperature: 1.0
    notes: "Strong general-purpose model with web search, temperature=1.0 only"

  gpt-5.1-codex-mini:
    litellm_model: openai/gpt-5.1-codex-mini
    aliases:
      - codex-mini
    constraints:
      temperature: 1.0
    notes: "Strong general-purpose model, temperature=1.0 only"

  gpt-5.1-codex:
    litellm_model: openai/gpt-5.1-codex
    aliases:
      - codex
    constraints:
      temperature: 1.0
    notes: "Strong general-purpose model, temperature=1.0 only"

  # ============================================
  # Azure OpenAI Models
  # ============================================
  # Note: LiteLLM reads AZURE_API_KEY, AZURE_API_BASE, and AZURE_API_VERSION from environment
  # AZURE_API_VERSION defaults to "2025-04-01-preview" if not set in .env
  azure-gpt-5-mini:
    litellm_model: azure/gpt-5-mini
    aliases:
      - azure-mini
      - az-mini
    constraints:
      temperature: 1.0  # gpt-5 family only supports 1.0
    notes: "Azure Mini deployment"

  # ============================================
  # Anthropic Models
  # ============================================

  claude-haiku-4.5:
    litellm_model: anthropic/claude-haiku-4-5-20251001
    aliases:
      - haiku
    notes: "Fast claude model"

  claude-sonnet-4.5:
    litellm_model: anthropic/claude-sonnet-4-5-20250929
    aliases:
      - sonnet
    notes: "Great for reasoning and analysis with native web search"

  claude-opus-4.5:
    litellm_model: anthropic/claude-opus-4-5-20251101
    aliases:
      - opus
    notes: "Best in class from anthropic"

  # ============================================
  # Google Models
  # ============================================

  gemini-2.5-pro:
    litellm_model: gemini/gemini-2.5-pro
    provider_web_search: true  # LiteLLM handles web search automatically
    aliases:
      - gemini-2.5
      - gemini-2
    notes: "High-quality Google model with grounding"

  gemini-3-flash:
    litellm_model: gemini/gemini-3-flash-preview
    provider_web_search: true
    aliases:
      - gemini-flash
      - flash
    notes: "Fastest Gemini 3 model"

  gemini-3-pro-preview:
    litellm_model: gemini/gemini-3-pro-preview
    provider_web_search: true  # LiteLLM handles web search automatically
    aliases:
      - gemini-3-pro
      - gemini-3
      - gemini
    notes: "Latest Google model with grounding"

  # ============================================
  # CLI Models
  # ============================================

  gemini-cli:
    provider: cli
    cli_command: gemini
    cli_args:
      - "-o"
      - "json"
      - "--approval-mode"
      - "auto_edit"
      # - "--yolo"  # Most dangerous: auto-approve all tools
    cli_env:
      GEMINI_API_KEY: "${GEMINI_API_KEY}"
    cli_parser: json
    aliases:
      - gem-cli
    notes: "Gemini CLI with auto-edit approval mode"

  codex-cli:
    provider: cli
    cli_command: codex
    cli_args:
      - "exec"
      - "--json"
      - "--full-auto"  # Low-friction sandboxed auto execution (safer than bypass)
      # - "--dangerously-bypass-approvals-and-sandbox"  # Most dangerous: bypass all approvals
    cli_env: {}
    cli_parser: jsonl
    aliases:
      - cx-cli
    notes: "Codex CLI with full-auto mode"

  claude-cli:
    provider: cli
    cli_command: claude
    cli_args:
      - "--print"  # Non-interactive mode
      - "--output-format"
      - "json"
      - "--permission-mode"
      - "acceptEdits"  # Auto-approve edits
      # - "--dangerously-skip-permissions"  # Most dangerous: bypass all permissions
    cli_env:
      ANTHROPIC_API_KEY: "${ANTHROPIC_API_KEY}"
      ANTHROPIC_MODEL: "claude-haiku-4-5-20251001"  # Use cheaper Haiku for integration tests
    cli_parser: json
    aliases:
      - cl-cli
    notes: "Claude CLI with acceptEdits permission mode, using Haiku for cost efficiency"

  # ============================================
  # AWS Bedrock Models
  # ============================================
  # To use AWS Bedrock models, configure AWS credentials in .env:
  #   AWS_ACCESS_KEY_ID=your_access_key
  #   AWS_SECRET_ACCESS_KEY=your_secret_key
  #   AWS_REGION_NAME=us-east-1

  bedrock-claude-4-5-sonnet:
    litellm_model: bedrock/anthropic.claude-sonnet-4-5-v2
    aliases:
      - bedrock-sonnet
    params:
      aws_region_name: us-east-1  # Override default region if needed
    notes: "Claude Sonnet 4.5 via AWS Bedrock"

  # ============================================
  # OpenRouter Models
  # ============================================

  # deepseek-chat:
  #   litellm_model: openrouter/deepseek/deepseek-chat-v3-0324
  #   aliases:
  #     - deepseek
  #   notes: "DeepSeek Chat via OpenRouter"

  # deepseek-r1:
  #   litellm_model: openrouter/deepseek/deepseek-r1
  #   aliases:
  #     - r1
  #   notes: "DeepSeek R1 reasoning model via OpenRouter"

  # qwen-coder:
  #   litellm_model: openrouter/qwen/qwen-2.5-coder-32b-instruct
  #   aliases:
  #     - qwen
  #   notes: "Qwen 2.5 Coder via OpenRouter"