models

type Cost struct {
	// Input is the cost per million input tokens
	Input float64

	// Output is the cost per million output tokens
	Output float64

	// CacheRead is the cost per million cached read tokens (optional)
	// Only applicable for models that support prompt caching
	CacheRead *float64

	// CacheWrite is the cost per million cached write tokens (optional)
	// Only applicable for models that support prompt caching
	CacheWrite *float64
}

type Limit struct {
	// Context is the maximum number of input tokens (context window size)
	Context int

	// Output is the maximum number of output tokens that can be generated
	Output int
}

type ModelInfo struct {
	// ID is the unique identifier for the model
	// Example: "claude-3-sonnet-20240620" or "gpt-4"
	ID string

	// Name is the human-readable name of the model
	// Example: "Claude 3 Sonnet" or "GPT-4"
	Name string

	// Attachment indicates whether the model supports file attachments
	Attachment bool

	// Reasoning indicates whether this is a reasoning/chain-of-thought model
	// Example: OpenAI's o1 models have this set to true
	Reasoning bool

	// Temperature indicates whether the model supports temperature parameter
	Temperature bool

	// Cost contains the pricing information for input/output tokens
	Cost Cost

	// Limit contains the context window and output token limits
	Limit Limit
}

type ModelsRegistry struct {
	// contains filtered or unexported fields
}

type OllamaLoadingResult struct {
	// Options contains the actual Ollama options used for loading
	// May differ from requested options if fallback occurred (e.g., CPU instead of GPU)
	Options *api.Options

	// Message describes the loading result
	// Example: "Model loaded successfully on GPU" or
	// "Insufficient GPU memory, falling back to CPU inference"
	Message string
}

type ProviderConfig struct {
	// ModelString specifies the model in the format "provider:model"
	// Example: "anthropic:claude-3-sonnet-20240620" or "openai:gpt-4"
	ModelString string

	// SystemPrompt sets the system message/instructions for the model
	SystemPrompt string

	// ProviderAPIKey is the API key for authentication with the provider
	// Used for OpenAI, Anthropic, Google, and Azure providers
	ProviderAPIKey string

	// ProviderURL is the base URL for the provider's API endpoint
	// Can be used to specify custom endpoints for OpenAI, Anthropic, Ollama, or Azure
	ProviderURL string

	// MaxTokens limits the maximum number of tokens in the response
	MaxTokens int

	// Temperature controls randomness in generation (0.0 to 1.0)
	// Lower values make output more focused and deterministic
	Temperature *float32

	// TopP implements nucleus sampling, controlling diversity
	// Value between 0.0 and 1.0, where 1.0 considers all tokens
	TopP *float32

	// TopK limits the number of tokens to sample from
	// Used by Anthropic and Google providers
	TopK *int32

	// StopSequences is a list of sequences that will stop generation
	StopSequences []string

	// NumGPU specifies the number of GPUs to use for inference
	NumGPU *int32

	// MainGPU specifies which GPU to use as the primary device
	MainGPU *int32

	// TLSSkipVerify skips TLS certificate verification (insecure)
	// Should only be used for development or with self-signed certificates
	TLSSkipVerify bool
}

type ProviderInfo struct {
	// ID is the unique identifier for the provider
	// Example: "anthropic", "openai", "google"
	ID string

	// Env lists the environment variables checked for API credentials
	// Example: ["ANTHROPIC_API_KEY"] or ["OPENAI_API_KEY"]
	Env []string

	// NPM is the NPM package name used by the provider (for reference)
	NPM string

	// Name is the human-readable name of the provider
	// Example: "Anthropic", "OpenAI", "Google"
	Name string

	// Models maps model IDs to their detailed information
	Models map[string]ModelInfo
}

type ProviderResult struct {
	// Model is the created LLM provider instance that implements
	// the ToolCallingChatModel interface for tool-enabled conversations
	Model model.ToolCallingChatModel

	// Message contains optional feedback for the user
	// Example: "Insufficient GPU memory, falling back to CPU inference"
	Message string
}