llm

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

type GatewayError struct {
	StatusCode int
	Body       string
	Headers    http.Header
	Retryable  bool
}

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

type LLMProvider interface {
	CompleteWithTools(ctx context.Context, req *ToolRequest) (*ToolResponse, error)
	Name() string
}

type Message struct {
	Role        string            `json:"role"`
	Content     string            `json:"content,omitempty"`
	Thinking    []ThinkingBlock   `json:"thinking,omitempty"`
	ToolUse     []ToolUseBlock    `json:"tool_use,omitempty"`
	ToolResults []ToolResultBlock `json:"tool_results,omitempty"`
}

type ModelLimits struct {
	ContextLimit  int // maximum prompt tokens
	OutputReserve int // tokens reserved for output generation
}

type StreamEvent struct {
	Type           string // "reasoning_delta", "content_delta", "tool_call_delta", "done"
	ContentDelta   string
	ReasoningDelta string
	ToolIndex      int
	ToolCallID     string
	ToolName       string
	ArgumentsDelta string
	FinishReason   string
	InputTokens    int
	OutputTokens   int
}

type StreamingLLMProvider interface {
	LLMProvider
	StreamCompleteWithTools(ctx context.Context, req *ToolRequest, callback func(StreamEvent)) (*ToolResponse, error)
}

type ThinkingBlock struct {
	Text      string `json:"text"`
	Signature string `json:"signature,omitempty"`
}

type TokenRateKey struct {
	Model     string
	TokenType string
}

type ToolCall struct {
	ID    string
	Name  string
	Input json.RawMessage
}

type ToolDefinition struct {
	Name        string          `json:"name"`
	Description string          `json:"description"`
	InputSchema json.RawMessage `json:"input_schema"`
}

type ToolRequest struct {
	Model       string
	System      string
	Messages    []Message
	Tools       []ToolDefinition
	MaxTokens   int
	Temperature float64
	// EnableReasoning turns on the provider's reasoning output — adaptive thinking on the native Anthropic path. Chat runs set it so the live thinking panel has content; other runs leave it off to avoid the extra thinking-token cost.
	EnableReasoning bool
	// ReasoningEffort, when set ("low"/"medium"/"high"), asks reasoning-capable models on the OpenAI-compatible path to emit reasoning. Ignored by the native Anthropic provider.
	ReasoningEffort string
}

type ToolResponse struct {
	Content      string
	Thinking     []ThinkingBlock // native reasoning blocks (Anthropic); empty for the OpenAI-compat path
	ToolCalls    []ToolCall
	InputTokens  int
	OutputTokens int
	StopReason   string // "end_turn", "tool_use", "max_tokens"
}

type ToolResultBlock struct {
	ToolUseID string `json:"tool_use_id"`
	Content   string `json:"content"`
	IsError   bool   `json:"is_error,omitempty"`
}

type ToolUseBlock struct {
	ID    string          `json:"id"`
	Name  string          `json:"name"`
	Input json.RawMessage `json:"input"`
}

type TruncationResult struct {
	Content      string
	WasTruncated bool
	FullLength   int
}