ragflow/internal/entity/models/openai.go

//
//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
//
//  Licensed under the Apache License, Version 2.0 (the "License");
//  you may not use this file except in compliance with the License.
//  You may obtain a copy of the License at
//
//      http://www.apache.org/licenses/LICENSE-2.0
//
//  Unless required by applicable law or agreed to in writing, software
//  distributed under the License is distributed on an "AS IS" BASIS,
//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
//  See the License for the specific language governing permissions and
//  limitations under the License.
//

package models

import (
	"bufio"
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
	"strings"
	"time"
)

// OpenAIModel implements ModelDriver for OpenAI (GPT models).
// The non-streaming call timeout is the shared nonStreamCallTimeout
// constant defined alongside the xAI driver in this package.
type OpenAIModel struct {
	BaseURL    map[string]string
	URLSuffix  URLSuffix
	httpClient *http.Client // Reusable HTTP client with connection pool
}

// NewOpenAIModel creates a new OpenAI model instance.
//
// We clone http.DefaultTransport so we keep Go's defaults for
// ProxyFromEnvironment, DialContext (with KeepAlive), HTTP/2,
// TLSHandshakeTimeout, and ExpectContinueTimeout, and only override
// the few connection-pool fields we care about.
//
// The Client itself has no Timeout. http.Client.Timeout would also
// cap the time spent reading the response body, which would cut off
// long-lived SSE streams in ChatStreamlyWithSender. Non-streaming
// callers wrap each request with context.WithTimeout instead.
func NewOpenAIModel(baseURL map[string]string, urlSuffix URLSuffix) *OpenAIModel {
	transport := http.DefaultTransport.(*http.Transport).Clone()
	transport.MaxIdleConns = 100
	transport.MaxIdleConnsPerHost = 10
	transport.IdleConnTimeout = 90 * time.Second
	transport.DisableCompression = false
	// Cap how long the client waits for the first response header.
	// This protects ChatStreamlyWithSender, which has no client-wide
	// timeout, against a server that opens the TCP connection and
	// then never sends a response.
	transport.ResponseHeaderTimeout = 60 * time.Second

	return &OpenAIModel{
		BaseURL:   baseURL,
		URLSuffix: urlSuffix,
		httpClient: &http.Client{
			Transport: transport,
		},
	}
}

func (z *OpenAIModel) NewInstance(baseURL map[string]string) ModelDriver {
	return NewOpenAIModel(baseURL, z.URLSuffix)
}

func (z *OpenAIModel) Name() string {
	return "openai"
}

// baseURLForRegion returns the base URL for the given region, or an
// error if no entry exists. This makes a misconfigured region fail
// fast with a clear message, instead of silently producing a relative
// URL that the HTTP transport then rejects.
func (z *OpenAIModel) baseURLForRegion(region string) (string, error) {
	base, ok := z.BaseURL[region]
	if !ok || base == "" {
		return "", fmt.Errorf("openai: no base URL configured for region %q", region)
	}
	return base, nil
}

// ChatWithMessages sends multiple messages with roles and returns the response
func (z *OpenAIModel) ChatWithMessages(modelName string, messages []Message, apiConfig *APIConfig, chatModelConfig *ChatConfig) (*ChatResponse, error) {
	if apiConfig == nil || apiConfig.ApiKey == nil || *apiConfig.ApiKey == "" {
		return nil, fmt.Errorf("api key is required")
	}

	if len(messages) == 0 {
		return nil, fmt.Errorf("messages is empty")
	}

	var region = "default"
	if apiConfig.Region != nil {
		region = *apiConfig.Region
	}

	baseURL, err := z.baseURLForRegion(region)
	if err != nil {
		return nil, err
	}
	url := fmt.Sprintf("%s/%s", baseURL, z.URLSuffix.Chat)

	// Convert messages to the format expected by the API
	apiMessages := make([]map[string]interface{}, len(messages))
	for i, msg := range messages {
		apiMessages[i] = map[string]interface{}{
			"role":    msg.Role,
			"content": msg.Content,
		}
	}

	// Build request body
	reqBody := map[string]interface{}{
		"model":       modelName,
		"messages":    apiMessages,
		"stream":      false,
		"temperature": 1,
	}

	// Note: do NOT propagate chatModelConfig.Stream into the request body
	// here. ChatWithMessages parses a single JSON response, so SSE/stream
	// must always be off for this code path.
	if chatModelConfig != nil {
		if chatModelConfig.MaxTokens != nil {
			reqBody["max_tokens"] = *chatModelConfig.MaxTokens
		}

		if chatModelConfig.Temperature != nil {
			reqBody["temperature"] = *chatModelConfig.Temperature
		}

		if chatModelConfig.TopP != nil {
			reqBody["top_p"] = *chatModelConfig.TopP
		}

		if chatModelConfig.Stop != nil {
			reqBody["stop"] = *chatModelConfig.Stop
		}
	}

	jsonData, err := json.Marshal(reqBody)
	if err != nil {
		return nil, fmt.Errorf("failed to marshal request: %w", err)
	}

	ctx, cancel := context.WithTimeout(context.Background(), nonStreamCallTimeout)
	defer cancel()

	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonData))
	if err != nil {
		return nil, fmt.Errorf("failed to create request: %w", err)
	}

	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", *apiConfig.ApiKey))

	resp, err := z.httpClient.Do(req)
	if err != nil {
		return nil, fmt.Errorf("failed to send request: %w", err)
	}
	defer resp.Body.Close()

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

	if resp.StatusCode != http.StatusOK {
		return nil, fmt.Errorf("API request failed with status %d: %s", resp.StatusCode, string(body))
	}

	// Parse response
	var result map[string]interface{}
	if err = json.Unmarshal(body, &result); err != nil {
		return nil, fmt.Errorf("failed to parse response: %w", err)
	}

	choices, ok := result["choices"].([]interface{})
	if !ok || len(choices) == 0 {
		return nil, fmt.Errorf("no choices in response")
	}

	firstChoice, ok := choices[0].(map[string]interface{})
	if !ok {
		return nil, fmt.Errorf("invalid choice format")
	}

	messageMap, ok := firstChoice["message"].(map[string]interface{})
	if !ok {
		return nil, fmt.Errorf("invalid message format")
	}

	content, ok := messageMap["content"].(string)
	if !ok {
		return nil, fmt.Errorf("invalid content format")
	}

	// OpenAI reasoning models (o-series and similar) return reasoning text in
	// the reasoning_content field. Pass it through when present.
	var reasonContent string
	if rc, ok := messageMap["reasoning_content"].(string); ok {
		reasonContent = rc
		if reasonContent != "" && reasonContent[0] == '\n' {
			reasonContent = reasonContent[1:]
		}
	}

	chatResponse := &ChatResponse{
		Answer:        &content,
		ReasonContent: &reasonContent,
	}

	return chatResponse, nil
}

// ChatStreamlyWithSender sends messages and streams the response via the
// sender function. Used for streaming chat responses with no extra channel.
func (z *OpenAIModel) ChatStreamlyWithSender(modelName string, messages []Message, apiConfig *APIConfig, chatModelConfig *ChatConfig, sender func(*string, *string) error) error {
	if len(messages) == 0 {
		return fmt.Errorf("messages is empty")
	}

	if apiConfig == nil || apiConfig.ApiKey == nil || *apiConfig.ApiKey == "" {
		return fmt.Errorf("api key is required")
	}

	var region = "default"
	if apiConfig.Region != nil && *apiConfig.Region != "" {
		region = *apiConfig.Region
	}

	baseURL, err := z.baseURLForRegion(region)
	if err != nil {
		return err
	}
	url := fmt.Sprintf("%s/%s", baseURL, z.URLSuffix.Chat)

	// Convert messages to API format (supports multimodal content)
	apiMessages := make([]map[string]interface{}, len(messages))
	for i, msg := range messages {
		apiMessages[i] = map[string]interface{}{
			"role":    msg.Role,
			"content": msg.Content,
		}
	}

	// Build request body with streaming on by default
	reqBody := map[string]interface{}{
		"model":    modelName,
		"messages": apiMessages,
		"stream":   true,
	}

	if chatModelConfig != nil {
		// Refuse to run if the caller explicitly asked for stream=false.
		// The body of this method only knows how to read SSE, so a non-SSE
		// JSON response would be parsed as if it were a stream and produce
		// no chunks. Better to fail clearly. Leave reqBody["stream"] as
		// the default (true) when Stream is nil or true.
		if chatModelConfig.Stream != nil && !*chatModelConfig.Stream {
			return fmt.Errorf("stream must be true in ChatStreamlyWithSender")
		}

		if chatModelConfig.MaxTokens != nil {
			reqBody["max_tokens"] = *chatModelConfig.MaxTokens
		}

		if chatModelConfig.Temperature != nil {
			reqBody["temperature"] = *chatModelConfig.Temperature
		}

		if chatModelConfig.TopP != nil {
			reqBody["top_p"] = *chatModelConfig.TopP
		}

		if chatModelConfig.Stop != nil {
			reqBody["stop"] = *chatModelConfig.Stop
		}
	}

	jsonData, err := json.Marshal(reqBody)
	if err != nil {
		return fmt.Errorf("failed to marshal request: %w", err)
	}

	// Use an explicit background context here so the request is at least
	// cancellable in principle. We do not attach a hard deadline because
	// SSE streams are long-lived. The transport's ResponseHeaderTimeout
	// caps the connection-establishment phase. Threading a real ctx
	// through the ModelDriver interface is a wider change for a follow-up.
	req, err := http.NewRequestWithContext(context.Background(), "POST", url, bytes.NewBuffer(jsonData))
	if err != nil {
		return fmt.Errorf("failed to create request: %w", err)
	}

	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", *apiConfig.ApiKey))

	resp, err := z.httpClient.Do(req)
	if err != nil {
		return fmt.Errorf("failed to send request: %w", err)
	}
	defer resp.Body.Close()

	if resp.StatusCode != http.StatusOK {
		body, _ := io.ReadAll(resp.Body)
		return fmt.Errorf("API request failed with status %d: %s", resp.StatusCode, string(body))
	}

	// SSE parsing: read line by line. The default bufio.Scanner buffer
	// is 64KB, which can be too small for long SSE chunks. Bump it to
	// 1MB so we never silently truncate a long data: line.
	scanner := bufio.NewScanner(resp.Body)
	scanner.Buffer(make([]byte, 64*1024), 1024*1024)
	// sawTerminal flips to true when the upstream actually told us the
	// stream is over (either a "[DONE]" marker or a non-empty
	// finish_reason). If the body closes before either of those, we
	// must not emit a synthetic "[DONE]" because that would hide a
	// truncated response from the caller.
	sawTerminal := false
	for scanner.Scan() {
		line := scanner.Text()

		// SSE data line starts with "data:"
		if !strings.HasPrefix(line, "data:") {
			continue
		}

		// Extract JSON after "data:"
		data := strings.TrimSpace(line[5:])

		// [DONE] marks the end of the stream
		if data == "[DONE]" {
			sawTerminal = true
			break
		}

		// Parse the JSON event
		var event map[string]interface{}
		if err = json.Unmarshal([]byte(data), &event); err != nil {
			continue
		}

		choices, ok := event["choices"].([]interface{})
		if !ok || len(choices) == 0 {
			continue
		}

		firstChoice, ok := choices[0].(map[string]interface{})
		if !ok {
			continue
		}

		delta, ok := firstChoice["delta"].(map[string]interface{})
		if !ok {
			continue
		}

		reasoningContent, ok := delta["reasoning_content"].(string)
		if ok && reasoningContent != "" {
			if err := sender(nil, &reasoningContent); err != nil {
				return err
			}
		}

		content, ok := delta["content"].(string)
		if ok && content != "" {
			if err := sender(&content, nil); err != nil {
				return err
			}
		}

		finishReason, ok := firstChoice["finish_reason"].(string)
		if ok && finishReason != "" {
			sawTerminal = true
			break
		}
	}

	if err := scanner.Err(); err != nil {
		return fmt.Errorf("failed to scan response body: %w", err)
	}
	if !sawTerminal {
		return fmt.Errorf("openai: stream ended before [DONE] or finish_reason")
	}

	// Send the [DONE] marker for OpenAI compatibility
	endOfStream := "[DONE]"
	if err := sender(&endOfStream, nil); err != nil {
		return err
	}

	return nil
}

// openaiEmbeddingResponse is the response shape returned by
// /v1/embeddings. The "index" field gives the position of the embedding
// in the input array, which we use to keep the output order stable
// even if the API returns items in a different order.
type openaiEmbeddingResponse struct {
	Data []struct {
		Index     int           `json:"index"`
		Embedding []interface{} `json:"embedding"`
	} `json:"data"`
}

// Encode turns a list of texts into embedding vectors using the
// OpenAI /v1/embeddings endpoint (e.g. text-embedding-3-small,
// text-embedding-3-large, text-embedding-ada-002). The output has
// one vector per input, in the same order the inputs were given.
func (z *OpenAIModel) Encode(modelName *string, texts []string, apiConfig *APIConfig, embeddingConfig *EmbeddingConfig) ([][]float64, error) {
	if len(texts) == 0 {
		return [][]float64{}, nil
	}

	if apiConfig == nil || apiConfig.ApiKey == nil || *apiConfig.ApiKey == "" {
		return nil, fmt.Errorf("api key is required")
	}

	if modelName == nil || *modelName == "" {
		return nil, fmt.Errorf("model name is required")
	}

	region := "default"
	if apiConfig.Region != nil && *apiConfig.Region != "" {
		region = *apiConfig.Region
	}

	baseURL, err := z.baseURLForRegion(region)
	if err != nil {
		return nil, err
	}
	url := fmt.Sprintf("%s/%s", baseURL, z.URLSuffix.Embedding)

	reqBody := map[string]interface{}{
		"model": *modelName,
		"input": texts,
	}
	if embeddingConfig != nil && embeddingConfig.Dimension > 0 {
		reqBody["dimensions"] = embeddingConfig.Dimension
	}

	jsonData, err := json.Marshal(reqBody)
	if err != nil {
		return nil, fmt.Errorf("failed to marshal request: %w", err)
	}

	ctx, cancel := context.WithTimeout(context.Background(), nonStreamCallTimeout)
	defer cancel()

	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewBuffer(jsonData))
	if err != nil {
		return nil, fmt.Errorf("failed to create request: %w", err)
	}

	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", *apiConfig.ApiKey))

	resp, err := z.httpClient.Do(req)
	if err != nil {
		return nil, fmt.Errorf("failed to send request: %w", err)
	}
	defer resp.Body.Close()

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

	if resp.StatusCode != http.StatusOK {
		return nil, fmt.Errorf("OpenAI embeddings API error: %s, body: %s", resp.Status, string(body))
	}

	var parsed openaiEmbeddingResponse
	if err = json.Unmarshal(body, &parsed); err != nil {
		return nil, fmt.Errorf("failed to parse response: %w", err)
	}

	embeddings := make([][]float64, len(texts))
	for _, item := range parsed.Data {
		if item.Index < 0 || item.Index >= len(texts) {
			continue
		}
		vec := make([]float64, len(item.Embedding))
		for j, v := range item.Embedding {
			switch val := v.(type) {
			case float64:
				vec[j] = val
			case float32:
				vec[j] = float64(val)
			default:
				return nil, fmt.Errorf("unexpected embedding value type at item %d index %d", item.Index, j)
			}
		}
		embeddings[item.Index] = vec
	}

	for i, vec := range embeddings {
		if vec == nil {
			return nil, fmt.Errorf("missing embedding for input at index %d", i)
		}
	}

	return embeddings, nil
}

// ListModels returns the list of model ids visible to the API key.
func (z *OpenAIModel) ListModels(apiConfig *APIConfig) ([]string, error) {
	if apiConfig == nil || apiConfig.ApiKey == nil || *apiConfig.ApiKey == "" {
		return nil, fmt.Errorf("api key is required")
	}

	var region = "default"
	if apiConfig.Region != nil {
		region = *apiConfig.Region
	}

	baseURL, err := z.baseURLForRegion(region)
	if err != nil {
		return nil, err
	}
	url := fmt.Sprintf("%s/%s", baseURL, z.URLSuffix.Models)

	ctx, cancel := context.WithTimeout(context.Background(), nonStreamCallTimeout)
	defer cancel()

	req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
	if err != nil {
		return nil, fmt.Errorf("failed to create request: %w", err)
	}

	// GET has no body, so Content-Type is not needed.
	req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", *apiConfig.ApiKey))

	resp, err := z.httpClient.Do(req)
	if err != nil {
		return nil, fmt.Errorf("failed to send request: %w", err)
	}
	defer resp.Body.Close()

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

	if resp.StatusCode != http.StatusOK {
		return nil, fmt.Errorf("API request failed with status %d: %s", resp.StatusCode, string(body))
	}

	// Parse response
	var result map[string]interface{}
	if err = json.Unmarshal(body, &result); err != nil {
		return nil, fmt.Errorf("failed to parse response: %w", err)
	}

	data, ok := result["data"].([]interface{})
	if !ok {
		return nil, fmt.Errorf("invalid models list format")
	}

	models := make([]string, 0)
	for _, model := range data {
		modelMap, ok := model.(map[string]interface{})
		if !ok {
			continue
		}
		modelName, ok := modelMap["id"].(string)
		if !ok {
			continue
		}
		models = append(models, modelName)
	}

	return models, nil
}

// Balance is not exposed by the OpenAI API, so this returns "no such method".
func (z *OpenAIModel) Balance(apiConfig *APIConfig) (map[string]interface{}, error) {
	return nil, fmt.Errorf("no such method")
}

// CheckConnection runs a lightweight ListModels call to verify the API key.
func (z *OpenAIModel) CheckConnection(apiConfig *APIConfig) error {
	_, err := z.ListModels(apiConfig)
	if err != nil {
		return err
	}
	return nil
}

// Rerank calculates similarity scores between query and documents. OpenAI does
// not expose a rerank API, so this is left unimplemented.
func (z *OpenAIModel) Rerank(modelName *string, query string, documents []string, apiConfig *APIConfig, rerankConfig *RerankConfig) (*RerankResponse, error) {
	return nil, fmt.Errorf("%s, Rerank not implemented", z.Name())
}