llm

type Client interface {
	Name() string
	Complete(ctx context.Context, req CompletionRequest) (CompletionResponse, error)
	// Stream delivers content deltas through onChunk as they arrive and
	// still returns the full accumulated response — "best-effort
	// incremental, full response always", so a caller that doesn't care
	// about incremental output can pass a nil onChunk and just use the
	// return value.
	Stream(ctx context.Context, req CompletionRequest, onChunk func(chunk string)) (CompletionResponse, error)
}

type CompletionRequest struct {
	Messages    []Message
	Model       string // empty uses the client's configured default
	Temperature *float64
	MaxTokens   *int
}

type CompletionResponse struct {
	Content      string
	Model        string
	InputTokens  int
	OutputTokens int
}

type Config struct {
	APIKey          string
	ChatModel       string
	BaseURL         string
	ReasoningEffort string
	// Headers carries extra request headers some gateways require or
	// recommend — e.g. OpenRouter's optional HTTP-Referer / X-Title.
	Headers map[string]string
	// HTTPClient is optional; nil falls back to a 5-minute-timeout client.
	HTTPClient *http.Client
}

type Message struct {
	Role    string // "system", "user", or "assistant"
	Content string
}