server/llm/openai/openai.go

package openai

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net/http"

	"github.com/iomodo/staff/llm"
)

// OpenAIProvider implements the LLMProvider interface for OpenAI
type OpenAIProvider struct {
	config llm.Config
	client *http.Client
}

// OpenAIRequest represents the OpenAI API request format
type OpenAIRequest struct {
	Model            string                `json:"model"`
	Messages         []OpenAIMessage       `json:"messages"`
	MaxTokens        *int                  `json:"max_tokens,omitempty"`
	Temperature      *float64              `json:"temperature,omitempty"`
	TopP             *float64              `json:"top_p,omitempty"`
	N                *int                  `json:"n,omitempty"`
	Stream           *bool                 `json:"stream,omitempty"`
	Stop             []string              `json:"stop,omitempty"`
	PresencePenalty  *float64              `json:"presence_penalty,omitempty"`
	FrequencyPenalty *float64              `json:"frequency_penalty,omitempty"`
	LogitBias        map[string]int        `json:"logit_bias,omitempty"`
	User             string                `json:"user,omitempty"`
	Tools            []OpenAITool          `json:"tools,omitempty"`
	ToolChoice       interface{}           `json:"tool_choice,omitempty"`
	ResponseFormat   *OpenAIResponseFormat `json:"response_format,omitempty"`
	Seed             *int64                `json:"seed,omitempty"`
}

// OpenAIMessage represents a message in OpenAI format
type OpenAIMessage struct {
	Role       string           `json:"role"`
	Content    string           `json:"content"`
	ToolCalls  []OpenAIToolCall `json:"tool_calls,omitempty"`
	ToolCallID string           `json:"tool_call_id,omitempty"`
	Name       string           `json:"name,omitempty"`
}

// OpenAIToolCall represents a tool call in OpenAI format
type OpenAIToolCall struct {
	ID       string         `json:"id"`
	Type     string         `json:"type"`
	Function OpenAIFunction `json:"function"`
}

// OpenAIFunction represents a function in OpenAI format
type OpenAIFunction struct {
	Name        string                 `json:"name"`
	Description string                 `json:"description,omitempty"`
	Parameters  map[string]interface{} `json:"parameters,omitempty"`
}

// OpenAITool represents a tool in OpenAI format
type OpenAITool struct {
	Type     string         `json:"type"`
	Function OpenAIFunction `json:"function"`
}

// OpenAIResponseFormat represents response format in OpenAI format
type OpenAIResponseFormat struct {
	Type string `json:"type"`
}

// OpenAIResponse represents the OpenAI API response format
type OpenAIResponse struct {
	ID                string         `json:"id"`
	Object            string         `json:"object"`
	Created           int64          `json:"created"`
	Model             string         `json:"model"`
	SystemFingerprint string         `json:"system_fingerprint,omitempty"`
	Choices           []OpenAIChoice `json:"choices"`
	Usage             OpenAIUsage    `json:"usage"`
}

// OpenAIChoice represents a choice in OpenAI response
type OpenAIChoice struct {
	Index        int             `json:"index"`
	Message      OpenAIMessage   `json:"message"`
	Logprobs     *OpenAILogprobs `json:"logprobs,omitempty"`
	FinishReason string          `json:"finish_reason"`
	Delta        *OpenAIMessage  `json:"delta,omitempty"`
}

// OpenAILogprobs represents log probabilities in OpenAI format
type OpenAILogprobs struct {
	Content []OpenAILogprobContent `json:"content,omitempty"`
}

// OpenAILogprobContent represents log probability content in OpenAI format
type OpenAILogprobContent struct {
	Token       string             `json:"token"`
	Logprob     float64            `json:"logprob"`
	Bytes       []int              `json:"bytes,omitempty"`
	TopLogprobs []OpenAITopLogprob `json:"top_logprobs,omitempty"`
}

// OpenAITopLogprob represents a top log probability in OpenAI format
type OpenAITopLogprob struct {
	Token   string  `json:"token"`
	Logprob float64 `json:"logprob"`
	Bytes   []int   `json:"bytes,omitempty"`
}

// OpenAIUsage represents usage information in OpenAI format
type OpenAIUsage struct {
	PromptTokens     int `json:"prompt_tokens"`
	CompletionTokens int `json:"completion_tokens"`
	TotalTokens      int `json:"total_tokens"`
}

// OpenAIEmbeddingRequest represents OpenAI embedding request
type OpenAIEmbeddingRequest struct {
	Input          interface{} `json:"input"`
	Model          string      `json:"model"`
	EncodingFormat string      `json:"encoding_format,omitempty"`
	Dimensions     *int        `json:"dimensions,omitempty"`
	User           string      `json:"user,omitempty"`
}

// OpenAIEmbeddingResponse represents OpenAI embedding response
type OpenAIEmbeddingResponse struct {
	Object string                `json:"object"`
	Data   []OpenAIEmbeddingData `json:"data"`
	Usage  OpenAIUsage           `json:"usage"`
	Model  string                `json:"model"`
}

// OpenAIEmbeddingData represents embedding data in OpenAI format
type OpenAIEmbeddingData struct {
	Object    string    `json:"object"`
	Embedding []float64 `json:"embedding"`
	Index     int       `json:"index"`
}

// OpenAIError represents an error from OpenAI API
type OpenAIError struct {
	Error struct {
		Message string `json:"message"`
		Type    string `json:"type"`
		Code    string `json:"code,omitempty"`
		Param   string `json:"param,omitempty"`
	} `json:"error"`
}

func New(config llm.Config) *OpenAIProvider {
	client := &http.Client{
		Timeout: config.Timeout,
	}

	return &OpenAIProvider{
		config: config,
		client: client,
	}
}

// ChatCompletion implements the LLMProvider interface for OpenAI
func (p *OpenAIProvider) ChatCompletion(ctx context.Context, req llm.ChatCompletionRequest) (*llm.ChatCompletionResponse, error) {
	// Convert our request to OpenAI format
	openAIReq := p.convertToOpenAIRequest(req)

	// Make the API call
	resp, err := p.makeOpenAIRequest(ctx, "/chat/completions", openAIReq)
	if err != nil {
		return nil, fmt.Errorf("OpenAI API request failed: %w", err)
	}

	// Parse the response
	var openAIResp OpenAIResponse
	if err := json.Unmarshal(resp, &openAIResp); err != nil {
		return nil, fmt.Errorf("failed to parse OpenAI response: %w", err)
	}

	// Convert back to our format
	return p.convertFromOpenAIResponse(openAIResp), nil
}

// CreateEmbeddings implements the LLMProvider interface for OpenAI
func (p *OpenAIProvider) CreateEmbeddings(ctx context.Context, req llm.EmbeddingRequest) (*llm.EmbeddingResponse, error) {
	// Convert our request to OpenAI format
	openAIReq := OpenAIEmbeddingRequest{
		Input:          req.Input,
		Model:          req.Model,
		EncodingFormat: req.EncodingFormat,
		Dimensions:     req.Dimensions,
		User:           req.User,
	}

	// Make the API call
	resp, err := p.makeOpenAIRequest(ctx, "/embeddings", openAIReq)
	if err != nil {
		return nil, fmt.Errorf("OpenAI embeddings API request failed: %w", err)
	}

	// Parse the response
	var openAIResp OpenAIEmbeddingResponse
	if err := json.Unmarshal(resp, &openAIResp); err != nil {
		return nil, fmt.Errorf("failed to parse OpenAI embeddings response: %w", err)
	}

	// Convert back to our format
	return p.convertFromOpenAIEmbeddingResponse(openAIResp), nil
}

// Close implements the LLMProvider interface
func (p *OpenAIProvider) Close() error {
	// Nothing to clean up for HTTP client
	return nil
}

// convertToOpenAIRequest converts our request format to OpenAI format
func (p *OpenAIProvider) convertToOpenAIRequest(req llm.ChatCompletionRequest) OpenAIRequest {
	openAIReq := OpenAIRequest{
		Model:            req.Model,
		MaxTokens:        req.MaxTokens,
		Temperature:      req.Temperature,
		TopP:             req.TopP,
		N:                req.N,
		Stream:           req.Stream,
		Stop:             req.Stop,
		PresencePenalty:  req.PresencePenalty,
		FrequencyPenalty: req.FrequencyPenalty,
		LogitBias:        req.LogitBias,
		User:             req.User,
		ToolChoice:       req.ToolChoice,
		Seed:             req.Seed,
	}

	// Convert messages
	openAIReq.Messages = make([]OpenAIMessage, len(req.Messages))
	for i, msg := range req.Messages {
		openAIReq.Messages[i] = OpenAIMessage{
			Role:       string(msg.Role),
			Content:    msg.Content,
			ToolCallID: msg.ToolCallID,
			Name:       msg.Name,
		}

		// Convert tool calls if present
		if len(msg.ToolCalls) > 0 {
			openAIReq.Messages[i].ToolCalls = make([]OpenAIToolCall, len(msg.ToolCalls))
			for j, toolCall := range msg.ToolCalls {
				openAIReq.Messages[i].ToolCalls[j] = OpenAIToolCall{
					ID:   toolCall.ID,
					Type: toolCall.Type,
					Function: OpenAIFunction{
						Name:        toolCall.Function.Name,
						Description: toolCall.Function.Description,
						Parameters:  toolCall.Function.Parameters,
					},
				}
			}
		}
	}

	// Convert tools if present
	if len(req.Tools) > 0 {
		openAIReq.Tools = make([]OpenAITool, len(req.Tools))
		for i, tool := range req.Tools {
			openAIReq.Tools[i] = OpenAITool{
				Type: tool.Type,
				Function: OpenAIFunction{
					Name:        tool.Function.Name,
					Description: tool.Function.Description,
					Parameters:  tool.Function.Parameters,
				},
			}
		}
	}

	// Convert response format if present
	if req.ResponseFormat != nil {
		openAIReq.ResponseFormat = &OpenAIResponseFormat{
			Type: req.ResponseFormat.Type,
		}
	}

	return openAIReq
}

// convertFromOpenAIResponse converts OpenAI response to our format
func (p *OpenAIProvider) convertFromOpenAIResponse(openAIResp OpenAIResponse) *llm.ChatCompletionResponse {
	resp := &llm.ChatCompletionResponse{
		ID:                openAIResp.ID,
		Object:            openAIResp.Object,
		Created:           openAIResp.Created,
		Model:             openAIResp.Model,
		SystemFingerprint: openAIResp.SystemFingerprint,
		Provider:          llm.ProviderOpenAI,
		Usage: llm.Usage{
			PromptTokens:     openAIResp.Usage.PromptTokens,
			CompletionTokens: openAIResp.Usage.CompletionTokens,
			TotalTokens:      openAIResp.Usage.TotalTokens,
		},
	}

	// Convert choices
	resp.Choices = make([]llm.ChatCompletionChoice, len(openAIResp.Choices))
	for i, choice := range openAIResp.Choices {
		resp.Choices[i] = llm.ChatCompletionChoice{
			Index:        choice.Index,
			FinishReason: choice.FinishReason,
			Message: llm.Message{
				Role:    llm.Role(choice.Message.Role),
				Content: choice.Message.Content,
				Name:    choice.Message.Name,
			},
		}

		// Convert tool calls if present
		if len(choice.Message.ToolCalls) > 0 {
			resp.Choices[i].Message.ToolCalls = make([]llm.ToolCall, len(choice.Message.ToolCalls))
			for j, toolCall := range choice.Message.ToolCalls {
				resp.Choices[i].Message.ToolCalls[j] = llm.ToolCall{
					ID:   toolCall.ID,
					Type: toolCall.Type,
					Function: llm.Function{
						Name:        toolCall.Function.Name,
						Description: toolCall.Function.Description,
						Parameters:  toolCall.Function.Parameters,
					},
				}
			}
		}

		// Convert logprobs if present
		if choice.Logprobs != nil {
			resp.Choices[i].Logprobs = &llm.Logprobs{
				Content: make([]llm.LogprobContent, len(choice.Logprobs.Content)),
			}
			for j, content := range choice.Logprobs.Content {
				resp.Choices[i].Logprobs.Content[j] = llm.LogprobContent{
					Token:   content.Token,
					Logprob: content.Logprob,
					Bytes:   content.Bytes,
				}
				if len(content.TopLogprobs) > 0 {
					resp.Choices[i].Logprobs.Content[j].TopLogprobs = make([]llm.TopLogprob, len(content.TopLogprobs))
					for k, topLogprob := range content.TopLogprobs {
						resp.Choices[i].Logprobs.Content[j].TopLogprobs[k] = llm.TopLogprob{
							Token:   topLogprob.Token,
							Logprob: topLogprob.Logprob,
							Bytes:   topLogprob.Bytes,
						}
					}
				}
			}
		}
	}

	return resp
}

// convertFromOpenAIEmbeddingResponse converts OpenAI embedding response to our format
func (p *OpenAIProvider) convertFromOpenAIEmbeddingResponse(openAIResp OpenAIEmbeddingResponse) *llm.EmbeddingResponse {
	resp := &llm.EmbeddingResponse{
		Object:   openAIResp.Object,
		Model:    openAIResp.Model,
		Provider: llm.ProviderOpenAI,
		Usage: llm.Usage{
			PromptTokens:     openAIResp.Usage.PromptTokens,
			CompletionTokens: openAIResp.Usage.CompletionTokens,
			TotalTokens:      openAIResp.Usage.TotalTokens,
		},
	}

	// Convert embedding data
	resp.Data = make([]llm.Embedding, len(openAIResp.Data))
	for i, data := range openAIResp.Data {
		resp.Data[i] = llm.Embedding{
			Object:    data.Object,
			Embedding: data.Embedding,
			Index:     data.Index,
		}
	}

	return resp
}

// makeOpenAIRequest makes an HTTP request to the OpenAI API
func (p *OpenAIProvider) makeOpenAIRequest(ctx context.Context, endpoint string, payload interface{}) ([]byte, error) {
	// Prepare request body
	jsonData, err := json.Marshal(payload)
	if err != nil {
		return nil, fmt.Errorf("failed to marshal request: %w", err)
	}

	// Create HTTP request
	url := p.config.BaseURL + endpoint
	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonData))
	if err != nil {
		return nil, fmt.Errorf("failed to create request: %w", err)
	}

	// Set headers
	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", "Bearer "+p.config.APIKey)

	// Add organization header if present
	if org, ok := p.config.ExtraConfig["organization"].(string); ok && org != "" {
		req.Header.Set("OpenAI-Organization", org)
	}

	// Make the request
	resp, err := p.client.Do(req)
	if err != nil {
		return nil, fmt.Errorf("HTTP request failed: %w", err)
	}
	defer resp.Body.Close()

	// Read response body
	body, err := io.ReadAll(resp.Body)
	if err != nil {
		return nil, fmt.Errorf("failed to read response body: %w", err)
	}

	// Check for errors
	if resp.StatusCode != http.StatusOK {
		var openAIErr OpenAIError
		if err := json.Unmarshal(body, &openAIErr); err != nil {
			return nil, fmt.Errorf("API error (status %d): %s", resp.StatusCode, string(body))
		}
		return nil, fmt.Errorf("OpenAI API error: %s (type: %s, code: %s)",
			openAIErr.Error.Message, openAIErr.Error.Type, openAIErr.Error.Code)
	}

	return body, nil
}