llm

type APIError struct {
	// Provider is the name of the provider that returned the error.
	Provider string
	// StatusCode is the HTTP status code returned.
	StatusCode int
	// Message is the error message from the API.
	Message string
	// Retryable indicates whether the request can be retried.
	Retryable bool
	// Err is the underlying error, if any.
	Err error
}

type ChatRequest struct {
	// Model is the identifier of the model to use.
	Model string
	// Messages is the session history to send to the model.
	Messages []Message
	// Temperature controls randomness in the response (0.0 to 2.0).
	// Lower values make output more deterministic.
	Temperature *float64
	// MaxTokens is the maximum number of tokens to generate.
	MaxTokens *int
	// TopP is the nucleus sampling parameter (0.0 to 1.0).
	TopP *float64
	// Stop is a list of sequences where the model will stop generating.
	Stop []string
	// Thinking enables extended thinking/reasoning mode.
	// Provider-specific handling in each provider implementation.
	Thinking *ThinkingRequest
	// Tools is a list of tools available for the model to call.
	Tools []Tool
	// ToolChoice controls how the model uses tools.
	// Values: "auto" (default), "required", "none", or a specific tool name.
	ToolChoice string
	// WebSearch enables provider-native web search when non-nil and Enabled.
	// Provider-specific handling in each provider implementation.
	WebSearch *WebSearchRequest
}

type ChatResponse struct {
	// Content is the generated text content.
	Content string
	// ReasoningContent holds the model's internal reasoning/thinking text.
	// Must be preserved and echoed back in the next assistant message when
	// reasoning is enabled, as required by some providers (e.g. Moonshot/KIMI).
	ReasoningContent string
	// FinishReason indicates why the model stopped generating.
	// Common values: "stop", "length", "content_filter", "tool_calls".
	FinishReason string
	// Usage contains token usage statistics.
	Usage Usage
	// ToolCalls contains tool calls requested by the model.
	// Only populated when FinishReason is "tool_calls".
	ToolCalls []ToolCall
}

type Config struct {
	// APIKey is the authentication key for the provider.
	APIKey string
	// AccountID carries provider-specific account metadata when required.
	AccountID string
	// BaseURL is the API endpoint URL. If empty, the provider's default is used.
	BaseURL string
	// OAuthCredentialProvider resolves dynamic bearer credentials on demand.
	OAuthCredentialProvider OAuthCredentialProvider
	// Timeout is the maximum time to wait for a response.
	// Default is 60 seconds if not specified unless DisableRequestTimeout is true.
	Timeout time.Duration
	// DisableRequestTimeout leaves the underlying request client without a
	// response timeout. Callers must cancel through the request context.
	DisableRequestTimeout bool

	// Retry configuration
	// MaxRetries is the maximum number of retry attempts for transient errors.
	// Default is 3 if not specified.
	MaxRetries int
	// InitialInterval is the initial backoff interval before the first retry.
	// Default is 1 second if not specified.
	InitialInterval time.Duration
	// MaxInterval is the maximum backoff interval between retries.
	// Default is 30 seconds if not specified.
	MaxInterval time.Duration
	// Multiplier is the factor by which the interval increases after each retry.
	// Default is 2.0 if not specified.
	Multiplier float64
}

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

type LogicalRetryConfig struct {
	MaxAttempts     int
	InitialInterval time.Duration
	MaxInterval     time.Duration
	Multiplier      float64
}

type Message struct {
	// Role identifies who sent the message (system, user, assistant, or tool).
	Role Role `json:"role"`
	// Content is the text content of the message.
	Content string `json:"content"`
	// ReasoningContent holds the model's internal reasoning/thinking text from a
	// previous assistant turn. Some providers (e.g. Moonshot/KIMI via OpenCode)
	// require this to be echoed back in subsequent turns when reasoning is enabled.
	ReasoningContent string `json:"reasoning_content,omitempty"`
	// Name is an optional identifier for the message sender.
	// Useful in multi-agent scenarios or for tool messages.
	Name string `json:"name,omitempty"`
	// ToolCallID is the ID of the tool call this message is responding to.
	// Required when Role is "tool".
	ToolCallID string `json:"tool_call_id,omitempty"`
	// ToolCalls contains tool calls made by the assistant.
	// Only set when Role is "assistant" and the model requests tool calls.
	ToolCalls []ToolCall `json:"tool_calls,omitempty"`
}

type OAuthCredential struct {
	AccessToken string
	AccountID   string
}

type OAuthCredentialProvider interface {
	ResolveCredential(ctx context.Context) (OAuthCredential, error)
}

type Option func(*Config)

type Provider interface {
	// Chat sends messages and returns the complete response.
	Chat(ctx context.Context, req *ChatRequest) (*ChatResponse, error)

	// ChatStream sends messages and streams the response.
	// Returns a channel that receives StreamEvents until Done is true or an error occurs.
	ChatStream(ctx context.Context, req *ChatRequest) (<-chan StreamEvent, error)

	// Name returns the provider name for logging and error messages.
	Name() string
}

type ProviderFactory func(cfg Config) (Provider, error)

type ProviderType string

type RequestOption func(*ChatRequest)

type Role string

type StreamEvent struct {
	// Delta is the incremental content received in this event.
	Delta string
	// Done indicates whether the stream has completed.
	Done bool
	// Error contains any error that occurred during streaming.
	Error error
	// Usage contains token usage statistics (only set when Done is true).
	Usage *Usage
	// ToolCalls contains tool calls accumulated during streaming.
	// Only populated when Done is true and the model requested tool calls.
	ToolCalls []ToolCall
	// FinishReason indicates why the model stopped (only set when Done is true).
	FinishReason string
}

type ThinkingEffort string

type ThinkingRequest struct {
	// Enabled activates thinking mode.
	Enabled bool
	// Effort controls reasoning depth: low, medium, high, xhigh.
	Effort ThinkingEffort
	// BudgetTokens sets explicit token budget (provider-specific).
	BudgetTokens *int
	// IncludeInOutput includes thinking blocks in the response content.
	IncludeInOutput bool
}

type Tool struct {
	// Type is always "function" for function tools.
	Type string `json:"type"`
	// Function contains the function definition.
	Function ToolFunction `json:"function"`
}

type ToolCall struct {
	// ID is a unique identifier for this tool call.
	ID string `json:"id"`
	// Type is always "function" for function calls.
	Type string `json:"type"`
	// Function contains the function call details.
	Function ToolCallFunction `json:"function"`
}

type ToolCallFunction struct {
	// Name is the name of the function to call.
	Name string `json:"name"`
	// Arguments is a JSON string containing the function arguments.
	Arguments string `json:"arguments"`
}

type ToolFunction struct {
	// Name is the function name (must match [a-zA-Z0-9_-]+).
	Name string `json:"name"`
	// Description explains what the function does.
	Description string `json:"description"`
	// Parameters is a JSON Schema describing the function parameters.
	Parameters map[string]any `json:"parameters"`
}

type Usage struct {
	// PromptTokens is the number of tokens in the prompt.
	PromptTokens int `json:"prompt_tokens"`
	// CompletionTokens is the number of tokens in the completion.
	CompletionTokens int `json:"completion_tokens"`
	// TotalTokens is the sum of prompt and completion tokens.
	TotalTokens int `json:"total_tokens"`
}

type UserLocation struct {
	// City is the city name.
	City string
	// Region is the region or state.
	Region string
	// Country is the ISO country code (e.g., "US").
	Country string
	// Timezone is the IANA timezone (e.g., "America/New_York").
	Timezone string
}

type WebSearchRequest struct {
	// Enabled activates provider-native web search.
	Enabled bool
	// MaxUses limits search invocations per request.
	// Anthropic: maps to max_uses. OpenRouter: maps to max_results.
	MaxUses *int
	// AllowedDomains restricts results to these domains (Anthropic only).
	AllowedDomains []string
	// BlockedDomains excludes results from these domains (Anthropic only).
	BlockedDomains []string
	// UserLocation localizes search results.
	UserLocation *UserLocation
}