From e540ec3a52811027656128041dec2ede8df3c30c Mon Sep 17 00:00:00 2001 From: jft0m <377632523@qq.com> Date: Tue, 14 Oct 2025 14:12:30 +0000 Subject: [PATCH 1/3] feat: add intelligent backend routing and model service MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Add modelService for centralized model management - Support dynamic model list from config file (data/supported_models.json) - Include 2025 latest models: GPT-4.1, o3, o4-mini, Gemini 2.5, etc. - File watcher for hot-reload configuration changes - Improve model detection logic in api.js - Priority: modelService lookup → prefix matching fallback - Smart backend routing based on model provider - Add intelligent routing endpoints - /v1/chat/completions: unified OpenAI-compatible endpoint - /v1/completions: legacy format support - Auto-route to Claude/OpenAI/Gemini based on requested model - Add Xcode system prompt support in openaiToClaude - Detect and preserve Xcode-specific system messages - Export handler functions for reuse - openaiClaudeRoutes: export handleChatCompletion - openaiRoutes: export handleResponses 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude --- src/app.js | 14 ++ src/routes/api.js | 257 +++++++++++++++++++++++++---- src/routes/openaiClaudeRoutes.js | 1 + src/routes/openaiRoutes.js | 1 + src/services/modelService.js | 266 +++++++++++++++++++++++++++++++ src/services/openaiToClaude.js | 19 ++- 6 files changed, 527 insertions(+), 31 deletions(-) create mode 100644 src/services/modelService.js diff --git a/src/app.js b/src/app.js index 42e8ce26..38b483d2 100644 --- a/src/app.js +++ b/src/app.js @@ -55,6 +55,11 @@ class Application { logger.info('🔄 Initializing pricing service...') await pricingService.initialize() + // 📋 初始化模型服务 + logger.info('🔄 Initializing model service...') + const modelService = require('./services/modelService') + await modelService.initialize() + // 📊 初始化缓存监控 await this.initializeCacheMonitoring() @@ -630,6 +635,15 @@ class Application { logger.error('❌ Error cleaning up pricing service:', error) } + // 清理 model service 的文件监听器 + try { + const modelService = require('./services/modelService') + modelService.cleanup() + logger.info('📋 Model service cleaned up') + } catch (error) { + logger.error('❌ Error cleaning up model service:', error) + } + // 停止限流清理服务 try { const rateLimitCleanupService = require('./services/rateLimitCleanupService') diff --git a/src/routes/api.js b/src/routes/api.js index f784cae6..805c298b 100644 --- a/src/routes/api.js +++ b/src/routes/api.js @@ -11,9 +11,138 @@ const logger = require('../utils/logger') const { getEffectiveModel, parseVendorPrefixedModel } = require('../utils/modelHelper') const sessionHelper = require('../utils/sessionHelper') const { updateRateLimitCounters } = require('../utils/rateLimitHelper') +const { handleChatCompletion } = require('./openaiClaudeRoutes') +const { + handleGenerateContent: geminiHandleGenerateContent, + handleStreamGenerateContent: geminiHandleStreamGenerateContent +} = require('./geminiRoutes') +const openaiRoutes = require('./openaiRoutes') const router = express.Router() +// 🔍 根据模型名称检测后端类型 +function detectBackendFromModel(modelName) { + if (!modelName) { + return 'claude' // 默认 Claude + } + + // 首先尝试使用 modelService 查找模型的 provider + try { + const modelService = require('../services/modelService') + const provider = modelService.getModelProvider(modelName) + + if (provider === 'anthropic') { + return 'claude' + } + if (provider === 'openai') { + return 'openai' + } + if (provider === 'google') { + return 'gemini' + } + } catch (error) { + logger.warn(`⚠️ Failed to detect backend from modelService: ${error.message}`) + } + + // 降级到前缀匹配作为后备方案 + const model = modelName.toLowerCase() + + // Claude 模型 + if (model.startsWith('claude-')) { + return 'claude' + } + + // OpenAI 模型 + if ( + model.startsWith('gpt-') || + model.startsWith('o1-') || + model.startsWith('o3-') || + model === 'chatgpt-4o-latest' + ) { + return 'openai' + } + + // Gemini 模型 + if (model.startsWith('gemini-')) { + return 'gemini' + } + + // 默认使用 Claude + return 'claude' +} + +// 🚀 智能后端路由处理器 +async function routeToBackend(req, res, requestedModel) { + const backend = detectBackendFromModel(requestedModel) + + logger.info(`🔀 Routing request - Model: ${requestedModel}, Backend: ${backend}`) + + // 检查权限 + const permissions = req.apiKey.permissions || 'all' + + if (backend === 'claude') { + // Claude 后端：通过 OpenAI 兼容层 + if (permissions !== 'all' && permissions !== 'claude') { + return res.status(403).json({ + error: { + message: 'This API key does not have permission to access Claude', + type: 'permission_denied', + code: 'permission_denied' + } + }) + } + await handleChatCompletion(req, res, req.apiKey) + } else if (backend === 'openai') { + // OpenAI 后端 + if (permissions !== 'all' && permissions !== 'openai') { + return res.status(403).json({ + error: { + message: 'This API key does not have permission to access OpenAI', + type: 'permission_denied', + code: 'permission_denied' + } + }) + } + return await openaiRoutes.handleResponses(req, res) + } else if (backend === 'gemini') { + // Gemini 后端 + if (permissions !== 'all' && permissions !== 'gemini') { + return res.status(403).json({ + error: { + message: 'This API key does not have permission to access Gemini', + type: 'permission_denied', + code: 'permission_denied' + } + }) + } + + // 转换为 Gemini 格式 + const geminiRequest = { + model: requestedModel, + messages: req.body.messages, + temperature: req.body.temperature || 0.7, + max_tokens: req.body.max_tokens || 4096, + stream: req.body.stream || false + } + + req.body = geminiRequest + + if (geminiRequest.stream) { + return await geminiHandleStreamGenerateContent(req, res) + } else { + return await geminiHandleGenerateContent(req, res) + } + } else { + return res.status(500).json({ + error: { + message: `Unsupported backend: ${backend}`, + type: 'server_error', + code: 'unsupported_backend' + } + }) + } +} + function queueRateLimitUpdate(rateLimitInfo, usageSummary, model, context = '') { if (!rateLimitInfo) { return Promise.resolve({ totalTokens: 0, totalCost: 0 }) @@ -722,40 +851,23 @@ router.post('/v1/messages', authenticateApiKey, handleMessagesRequest) // 🚀 Claude API messages 端点 - /claude/v1/messages (别名) router.post('/claude/v1/messages', authenticateApiKey, handleMessagesRequest) -// 📋 模型列表端点 - Claude Code 客户端需要 +// 📋 模型列表端点 - 支持 Claude, OpenAI, Gemini router.get('/v1/models', authenticateApiKey, async (req, res) => { try { - // 返回支持的模型列表 - const models = [ - { - id: 'claude-3-5-sonnet-20241022', - object: 'model', - created: 1669599635, - owned_by: 'anthropic' - }, - { - id: 'claude-3-5-haiku-20241022', - object: 'model', - created: 1669599635, - owned_by: 'anthropic' - }, - { - id: 'claude-3-opus-20240229', - object: 'model', - created: 1669599635, - owned_by: 'anthropic' - }, - { - id: 'claude-sonnet-4-20250514', - object: 'model', - created: 1669599635, - owned_by: 'anthropic' - } - ] + const modelService = require('../services/modelService') + + // 从 modelService 获取所有支持的模型 + const models = modelService.getAllModels() + + // 可选：根据 API Key 的模型限制过滤 + let filteredModels = models + if (req.apiKey.enableModelRestriction && req.apiKey.restrictedModels?.length > 0) { + filteredModels = models.filter((model) => req.apiKey.restrictedModels.includes(model.id)) + } res.json({ object: 'list', - data: models + data: filteredModels }) } catch (error) { logger.error('❌ Models list error:', error) @@ -766,6 +878,93 @@ router.get('/v1/models', authenticateApiKey, async (req, res) => { } }) +// 🔄 OpenAI 兼容的 chat/completions 端点（智能后端路由） +router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => { + try { + // 验证必需参数 + if (!req.body.messages || !Array.isArray(req.body.messages) || req.body.messages.length === 0) { + return res.status(400).json({ + error: { + message: 'Messages array is required and cannot be empty', + type: 'invalid_request_error', + code: 'invalid_request' + } + }) + } + + const requestedModel = req.body.model || 'claude-3-5-sonnet-20241022' + req.body.model = requestedModel // 确保模型已设置 + + // 使用统一的后端路由处理器 + await routeToBackend(req, res, requestedModel) + } catch (error) { + logger.error('❌ OpenAI chat/completions error:', error) + if (!res.headersSent) { + res.status(500).json({ + error: { + message: 'Internal server error', + type: 'server_error', + code: 'internal_error' + } + }) + } + } +}) + +// 🔄 OpenAI 兼容的 completions 端点（传统格式，智能后端路由） +router.post('/v1/completions', authenticateApiKey, async (req, res) => { + try { + // 验证必需参数 + if (!req.body.prompt) { + return res.status(400).json({ + error: { + message: 'Prompt is required', + type: 'invalid_request_error', + code: 'invalid_request' + } + }) + } + + // 将传统 completions 格式转换为 chat 格式 + const originalBody = req.body + const requestedModel = originalBody.model || 'claude-3-5-sonnet-20241022' + + req.body = { + model: requestedModel, + messages: [ + { + role: 'user', + content: originalBody.prompt + } + ], + max_tokens: originalBody.max_tokens, + temperature: originalBody.temperature, + top_p: originalBody.top_p, + stream: originalBody.stream, + stop: originalBody.stop, + n: originalBody.n || 1, + presence_penalty: originalBody.presence_penalty, + frequency_penalty: originalBody.frequency_penalty, + logit_bias: originalBody.logit_bias, + user: originalBody.user + } + + // 使用统一的后端路由处理器 + await routeToBackend(req, res, requestedModel) + } catch (error) { + logger.error('❌ OpenAI completions error:', error) + if (!res.headersSent) { + res.status(500).json({ + error: { + message: 'Failed to process completion request', + type: 'server_error', + code: 'internal_error' + } + }) + } + } +}) + // 🏥 健康检查端点 router.get('/health', async (req, res) => { try { diff --git a/src/routes/openaiClaudeRoutes.js b/src/routes/openaiClaudeRoutes.js index f5db5665..87a9d38f 100644 --- a/src/routes/openaiClaudeRoutes.js +++ b/src/routes/openaiClaudeRoutes.js @@ -490,3 +490,4 @@ router.post('/v1/completions', authenticateApiKey, async (req, res) => { }) module.exports = router +module.exports.handleChatCompletion = handleChatCompletion diff --git a/src/routes/openaiRoutes.js b/src/routes/openaiRoutes.js index 13776c8d..bfebcd0f 100644 --- a/src/routes/openaiRoutes.js +++ b/src/routes/openaiRoutes.js @@ -919,3 +919,4 @@ router.get('/key-info', authenticateApiKey, async (req, res) => { }) module.exports = router +module.exports.handleResponses = handleResponses diff --git a/src/services/modelService.js b/src/services/modelService.js new file mode 100644 index 00000000..1bee9dba --- /dev/null +++ b/src/services/modelService.js @@ -0,0 +1,266 @@ +const fs = require('fs') +const path = require('path') +const logger = require('../utils/logger') + +/** + * 模型服务 + * 管理系统支持的 AI 模型列表 + * 与 pricingService 独立，专注于"支持哪些模型"而不是"如何计费" + */ +class ModelService { + constructor() { + this.modelsFile = path.join(process.cwd(), 'data', 'supported_models.json') + this.supportedModels = null + this.fileWatcher = null + } + + /** + * 初始化模型服务 + */ + async initialize() { + try { + this.loadModels() + this.setupFileWatcher() + logger.success('✅ Model service initialized successfully') + } catch (error) { + logger.error('❌ Failed to initialize model service:', error) + } + } + + /** + * 加载支持的模型配置 + */ + loadModels() { + try { + if (fs.existsSync(this.modelsFile)) { + const data = fs.readFileSync(this.modelsFile, 'utf8') + this.supportedModels = JSON.parse(data) + + const totalModels = Object.values(this.supportedModels).reduce( + (sum, config) => sum + config.models.length, + 0 + ) + + logger.info(`📋 Loaded ${totalModels} supported models from configuration`) + } else { + logger.warn('⚠️ Supported models file not found, using defaults') + this.supportedModels = this.getDefaultModels() + + // 创建默认配置文件 + this.saveDefaultConfig() + } + } catch (error) { + logger.error('❌ Failed to load supported models:', error) + this.supportedModels = this.getDefaultModels() + } + } + + /** + * 获取默认模型配置（后备方案） + */ + getDefaultModels() { + return { + claude: { + provider: 'anthropic', + description: 'Claude models from Anthropic', + models: [ + 'claude-sonnet-4-5-20250929', + 'claude-opus-4-1-20250805', + 'claude-sonnet-4-20250514', + 'claude-opus-4-20250514', + 'claude-3-7-sonnet-20250219', + 'claude-3-5-sonnet-20241022', + 'claude-3-5-haiku-20241022', + 'claude-3-opus-20240229', + 'claude-3-haiku-20240307' + ] + }, + openai: { + provider: 'openai', + description: 'OpenAI GPT models', + models: [ + 'gpt-4o', + 'gpt-4o-mini', + 'gpt-4.1', + 'gpt-4.1-mini', + 'gpt-4.1-nano', + 'gpt-4-turbo', + 'gpt-4', + 'gpt-3.5-turbo', + 'o3', + 'o4-mini', + 'chatgpt-4o-latest' + ] + }, + gemini: { + provider: 'google', + description: 'Google Gemini models', + models: [ + 'gemini-1.5-pro', + 'gemini-1.5-flash', + 'gemini-2.0-flash', + 'gemini-2.0-flash-exp', + 'gemini-2.0-flash-thinking', + 'gemini-2.0-flash-thinking-exp', + 'gemini-2.0-pro', + 'gemini-2.5-flash', + 'gemini-2.5-flash-lite', + 'gemini-2.5-pro' + ] + } + } + } + + /** + * 保存默认配置到文件 + */ + saveDefaultConfig() { + try { + const dataDir = path.dirname(this.modelsFile) + if (!fs.existsSync(dataDir)) { + fs.mkdirSync(dataDir, { recursive: true }) + } + + fs.writeFileSync(this.modelsFile, JSON.stringify(this.supportedModels, null, 2)) + logger.info('💾 Created default supported_models.json configuration') + } catch (error) { + logger.error('❌ Failed to save default config:', error) + } + } + + /** + * 获取所有支持的模型（OpenAI API 格式） + */ + getAllModels() { + const models = [] + const now = Math.floor(Date.now() / 1000) + + for (const [_service, config] of Object.entries(this.supportedModels)) { + for (const modelId of config.models) { + models.push({ + id: modelId, + object: 'model', + created: now, + owned_by: config.provider + }) + } + } + + return models.sort((a, b) => { + // 先按 provider 排序，再按 model id 排序 + if (a.owned_by !== b.owned_by) { + return a.owned_by.localeCompare(b.owned_by) + } + return a.id.localeCompare(b.id) + }) + } + + /** + * 按 provider 获取模型 + * @param {string} provider - 'anthropic', 'openai', 'google' 等 + */ + getModelsByProvider(provider) { + return this.getAllModels().filter((m) => m.owned_by === provider) + } + + /** + * 检查模型是否被支持 + * @param {string} modelId - 模型 ID + */ + isModelSupported(modelId) { + if (!modelId) { + return false + } + return this.getAllModels().some((m) => m.id === modelId) + } + + /** + * 获取模型的 provider + * @param {string} modelId - 模型 ID + */ + getModelProvider(modelId) { + const model = this.getAllModels().find((m) => m.id === modelId) + return model ? model.owned_by : null + } + + /** + * 重新加载模型配置 + */ + reloadModels() { + logger.info('🔄 Reloading supported models configuration...') + this.loadModels() + } + + /** + * 设置文件监听器（监听配置文件变化） + */ + setupFileWatcher() { + try { + // 如果已有监听器，先关闭 + if (this.fileWatcher) { + this.fileWatcher.close() + this.fileWatcher = null + } + + // 只有文件存在时才设置监听器 + if (!fs.existsSync(this.modelsFile)) { + logger.debug('📋 Models file does not exist yet, skipping file watcher setup') + return + } + + // 使用 fs.watchFile 监听文件变化 + const watchOptions = { + persistent: true, + interval: 60000 // 每60秒检查一次 + } + + let lastMtime = fs.statSync(this.modelsFile).mtimeMs + + fs.watchFile(this.modelsFile, watchOptions, (curr, _prev) => { + if (curr.mtimeMs !== lastMtime) { + lastMtime = curr.mtimeMs + logger.info('📋 Detected change in supported_models.json, reloading...') + this.reloadModels() + } + }) + + // 保存引用以便清理 + this.fileWatcher = { + close: () => fs.unwatchFile(this.modelsFile) + } + + logger.info('👁️ File watcher set up for supported_models.json') + } catch (error) { + logger.error('❌ Failed to setup file watcher:', error) + } + } + + /** + * 获取服务状态 + */ + getStatus() { + const totalModels = this.supportedModels + ? Object.values(this.supportedModels).reduce((sum, config) => sum + config.models.length, 0) + : 0 + + return { + initialized: this.supportedModels !== null, + totalModels, + providers: this.supportedModels ? Object.keys(this.supportedModels) : [], + fileExists: fs.existsSync(this.modelsFile) + } + } + + /** + * 清理资源 + */ + cleanup() { + if (this.fileWatcher) { + this.fileWatcher.close() + this.fileWatcher = null + logger.debug('📋 Model service file watcher closed') + } + } +} + +module.exports = new ModelService() diff --git a/src/services/openaiToClaude.js b/src/services/openaiToClaude.js index 10c8ae24..1f335f0e 100644 --- a/src/services/openaiToClaude.js +++ b/src/services/openaiToClaude.js @@ -31,10 +31,25 @@ class OpenAIToClaudeConverter { stream: openaiRequest.stream || false } - // Claude Code 必需的系统消息 + // 定义 Claude Code 的默认系统提示词 const claudeCodeSystemMessage = "You are Claude Code, Anthropic's official CLI for Claude." - claudeRequest.system = claudeCodeSystemMessage + // 如果 OpenAI 请求中包含系统消息,提取并检查 + const systemMessage = this._extractSystemMessage(openaiRequest.messages) + if (systemMessage && systemMessage.includes('You are currently in Xcode')) { + // Xcode 系统提示词 + claudeRequest.system = systemMessage + logger.info( + `🔍 Xcode request detected, using Xcode system prompt (${systemMessage.length} chars)` + ) + logger.debug(`📋 System prompt preview: ${systemMessage.substring(0, 150)}...`) + } else { + // 使用 Claude Code 默认系统提示词 + claudeRequest.system = claudeCodeSystemMessage + logger.debug( + `📋 Using Claude Code default system prompt${systemMessage ? ' (ignored custom prompt)' : ''}` + ) + } // 处理停止序列 if (openaiRequest.stop) { From 344599f318d525e50adb65d01afc2c58f2153baa Mon Sep 17 00:00:00 2001 From: jft0m <377632523@qq.com> Date: Tue, 14 Oct 2025 14:30:23 +0000 Subject: [PATCH 2/3] refactor: extract intelligent routing to unified.js MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Created new src/routes/unified.js (225 lines) - detectBackendFromModel(): Detects backend from model name - routeToBackend(): Routes to Claude/OpenAI/Gemini with permission checks - POST /v1/chat/completions: OpenAI-compatible endpoint with intelligent routing - POST /v1/completions: Legacy completions endpoint with intelligent routing - Updated src/routes/api.js (reduced from 1185 to 968 lines) - Removed ~217 lines of routing logic - Kept Claude-specific endpoints (/api/v1/messages) - Maintained all other Claude API functionality - Updated src/app.js - Added unifiedRoutes registration at /api prefix Benefits: - Single responsibility: api.js focuses on Claude API routes - Better organization: routing logic isolated in unified.js - Easier maintenance: changes to routing won't affect Claude code - File size reduction: api.js reduced by 18% Tested: - ✅ Claude model routing via /v1/chat/completions - ✅ OpenAI model routing (correct backend detection) - ✅ Gemini model routing (correct backend detection) - ✅ Legacy /v1/completions endpoint - ✅ All tests pass, no regressions 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude --- src/app.js | 2 + src/routes/api.js | 217 ---------------------------------------- src/routes/unified.js | 225 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 227 insertions(+), 217 deletions(-) create mode 100644 src/routes/unified.js diff --git a/src/app.js b/src/app.js index 38b483d2..4f86c6cf 100644 --- a/src/app.js +++ b/src/app.js @@ -14,6 +14,7 @@ const cacheMonitor = require('./utils/cacheMonitor') // Import routes const apiRoutes = require('./routes/api') +const unifiedRoutes = require('./routes/unified') const adminRoutes = require('./routes/admin') const webRoutes = require('./routes/web') const apiStatsRoutes = require('./routes/apiStats') @@ -256,6 +257,7 @@ class Application { // 🛣️ 路由 this.app.use('/api', apiRoutes) + this.app.use('/api', unifiedRoutes) // 统一智能路由（支持 /v1/chat/completions 等） this.app.use('/claude', apiRoutes) // /claude 路由别名，与 /api 功能相同 this.app.use('/admin', adminRoutes) this.app.use('/users', userRoutes) diff --git a/src/routes/api.js b/src/routes/api.js index 805c298b..2d365440 100644 --- a/src/routes/api.js +++ b/src/routes/api.js @@ -11,138 +11,8 @@ const logger = require('../utils/logger') const { getEffectiveModel, parseVendorPrefixedModel } = require('../utils/modelHelper') const sessionHelper = require('../utils/sessionHelper') const { updateRateLimitCounters } = require('../utils/rateLimitHelper') -const { handleChatCompletion } = require('./openaiClaudeRoutes') -const { - handleGenerateContent: geminiHandleGenerateContent, - handleStreamGenerateContent: geminiHandleStreamGenerateContent -} = require('./geminiRoutes') -const openaiRoutes = require('./openaiRoutes') - const router = express.Router() -// 🔍 根据模型名称检测后端类型 -function detectBackendFromModel(modelName) { - if (!modelName) { - return 'claude' // 默认 Claude - } - - // 首先尝试使用 modelService 查找模型的 provider - try { - const modelService = require('../services/modelService') - const provider = modelService.getModelProvider(modelName) - - if (provider === 'anthropic') { - return 'claude' - } - if (provider === 'openai') { - return 'openai' - } - if (provider === 'google') { - return 'gemini' - } - } catch (error) { - logger.warn(`⚠️ Failed to detect backend from modelService: ${error.message}`) - } - - // 降级到前缀匹配作为后备方案 - const model = modelName.toLowerCase() - - // Claude 模型 - if (model.startsWith('claude-')) { - return 'claude' - } - - // OpenAI 模型 - if ( - model.startsWith('gpt-') || - model.startsWith('o1-') || - model.startsWith('o3-') || - model === 'chatgpt-4o-latest' - ) { - return 'openai' - } - - // Gemini 模型 - if (model.startsWith('gemini-')) { - return 'gemini' - } - - // 默认使用 Claude - return 'claude' -} - -// 🚀 智能后端路由处理器 -async function routeToBackend(req, res, requestedModel) { - const backend = detectBackendFromModel(requestedModel) - - logger.info(`🔀 Routing request - Model: ${requestedModel}, Backend: ${backend}`) - - // 检查权限 - const permissions = req.apiKey.permissions || 'all' - - if (backend === 'claude') { - // Claude 后端：通过 OpenAI 兼容层 - if (permissions !== 'all' && permissions !== 'claude') { - return res.status(403).json({ - error: { - message: 'This API key does not have permission to access Claude', - type: 'permission_denied', - code: 'permission_denied' - } - }) - } - await handleChatCompletion(req, res, req.apiKey) - } else if (backend === 'openai') { - // OpenAI 后端 - if (permissions !== 'all' && permissions !== 'openai') { - return res.status(403).json({ - error: { - message: 'This API key does not have permission to access OpenAI', - type: 'permission_denied', - code: 'permission_denied' - } - }) - } - return await openaiRoutes.handleResponses(req, res) - } else if (backend === 'gemini') { - // Gemini 后端 - if (permissions !== 'all' && permissions !== 'gemini') { - return res.status(403).json({ - error: { - message: 'This API key does not have permission to access Gemini', - type: 'permission_denied', - code: 'permission_denied' - } - }) - } - - // 转换为 Gemini 格式 - const geminiRequest = { - model: requestedModel, - messages: req.body.messages, - temperature: req.body.temperature || 0.7, - max_tokens: req.body.max_tokens || 4096, - stream: req.body.stream || false - } - - req.body = geminiRequest - - if (geminiRequest.stream) { - return await geminiHandleStreamGenerateContent(req, res) - } else { - return await geminiHandleGenerateContent(req, res) - } - } else { - return res.status(500).json({ - error: { - message: `Unsupported backend: ${backend}`, - type: 'server_error', - code: 'unsupported_backend' - } - }) - } -} - function queueRateLimitUpdate(rateLimitInfo, usageSummary, model, context = '') { if (!rateLimitInfo) { return Promise.resolve({ totalTokens: 0, totalCost: 0 }) @@ -878,93 +748,6 @@ router.get('/v1/models', authenticateApiKey, async (req, res) => { } }) -// 🔄 OpenAI 兼容的 chat/completions 端点（智能后端路由） -router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => { - try { - // 验证必需参数 - if (!req.body.messages || !Array.isArray(req.body.messages) || req.body.messages.length === 0) { - return res.status(400).json({ - error: { - message: 'Messages array is required and cannot be empty', - type: 'invalid_request_error', - code: 'invalid_request' - } - }) - } - - const requestedModel = req.body.model || 'claude-3-5-sonnet-20241022' - req.body.model = requestedModel // 确保模型已设置 - - // 使用统一的后端路由处理器 - await routeToBackend(req, res, requestedModel) - } catch (error) { - logger.error('❌ OpenAI chat/completions error:', error) - if (!res.headersSent) { - res.status(500).json({ - error: { - message: 'Internal server error', - type: 'server_error', - code: 'internal_error' - } - }) - } - } -}) - -// 🔄 OpenAI 兼容的 completions 端点（传统格式，智能后端路由） -router.post('/v1/completions', authenticateApiKey, async (req, res) => { - try { - // 验证必需参数 - if (!req.body.prompt) { - return res.status(400).json({ - error: { - message: 'Prompt is required', - type: 'invalid_request_error', - code: 'invalid_request' - } - }) - } - - // 将传统 completions 格式转换为 chat 格式 - const originalBody = req.body - const requestedModel = originalBody.model || 'claude-3-5-sonnet-20241022' - - req.body = { - model: requestedModel, - messages: [ - { - role: 'user', - content: originalBody.prompt - } - ], - max_tokens: originalBody.max_tokens, - temperature: originalBody.temperature, - top_p: originalBody.top_p, - stream: originalBody.stream, - stop: originalBody.stop, - n: originalBody.n || 1, - presence_penalty: originalBody.presence_penalty, - frequency_penalty: originalBody.frequency_penalty, - logit_bias: originalBody.logit_bias, - user: originalBody.user - } - - // 使用统一的后端路由处理器 - await routeToBackend(req, res, requestedModel) - } catch (error) { - logger.error('❌ OpenAI completions error:', error) - if (!res.headersSent) { - res.status(500).json({ - error: { - message: 'Failed to process completion request', - type: 'server_error', - code: 'internal_error' - } - }) - } - } -}) - // 🏥 健康检查端点 router.get('/health', async (req, res) => { try { diff --git a/src/routes/unified.js b/src/routes/unified.js new file mode 100644 index 00000000..c1f320d2 --- /dev/null +++ b/src/routes/unified.js @@ -0,0 +1,225 @@ +const express = require('express') +const { authenticateApiKey } = require('../middleware/auth') +const logger = require('../utils/logger') +const { handleChatCompletion } = require('./openaiClaudeRoutes') +const { + handleGenerateContent: geminiHandleGenerateContent, + handleStreamGenerateContent: geminiHandleStreamGenerateContent +} = require('./geminiRoutes') +const openaiRoutes = require('./openaiRoutes') + +const router = express.Router() + +// 🔍 根据模型名称检测后端类型 +function detectBackendFromModel(modelName) { + if (!modelName) { + return 'claude' // 默认 Claude + } + + // 首先尝试使用 modelService 查找模型的 provider + try { + const modelService = require('../services/modelService') + const provider = modelService.getModelProvider(modelName) + + if (provider === 'anthropic') { + return 'claude' + } + if (provider === 'openai') { + return 'openai' + } + if (provider === 'google') { + return 'gemini' + } + } catch (error) { + logger.warn(`⚠️ Failed to detect backend from modelService: ${error.message}`) + } + + // 降级到前缀匹配作为后备方案 + const model = modelName.toLowerCase() + + // Claude 模型 + if (model.startsWith('claude-')) { + return 'claude' + } + + // OpenAI 模型 + if ( + model.startsWith('gpt-') || + model.startsWith('o1-') || + model.startsWith('o3-') || + model === 'chatgpt-4o-latest' + ) { + return 'openai' + } + + // Gemini 模型 + if (model.startsWith('gemini-')) { + return 'gemini' + } + + // 默认使用 Claude + return 'claude' +} + +// 🚀 智能后端路由处理器 +async function routeToBackend(req, res, requestedModel) { + const backend = detectBackendFromModel(requestedModel) + + logger.info(`🔀 Routing request - Model: ${requestedModel}, Backend: ${backend}`) + + // 检查权限 + const permissions = req.apiKey.permissions || 'all' + + if (backend === 'claude') { + // Claude 后端：通过 OpenAI 兼容层 + if (permissions !== 'all' && permissions !== 'claude') { + return res.status(403).json({ + error: { + message: 'This API key does not have permission to access Claude', + type: 'permission_denied', + code: 'permission_denied' + } + }) + } + await handleChatCompletion(req, res, req.apiKey) + } else if (backend === 'openai') { + // OpenAI 后端 + if (permissions !== 'all' && permissions !== 'openai') { + return res.status(403).json({ + error: { + message: 'This API key does not have permission to access OpenAI', + type: 'permission_denied', + code: 'permission_denied' + } + }) + } + return await openaiRoutes.handleResponses(req, res) + } else if (backend === 'gemini') { + // Gemini 后端 + if (permissions !== 'all' && permissions !== 'gemini') { + return res.status(403).json({ + error: { + message: 'This API key does not have permission to access Gemini', + type: 'permission_denied', + code: 'permission_denied' + } + }) + } + + // 转换为 Gemini 格式 + const geminiRequest = { + model: requestedModel, + messages: req.body.messages, + temperature: req.body.temperature || 0.7, + max_tokens: req.body.max_tokens || 4096, + stream: req.body.stream || false + } + + req.body = geminiRequest + + if (geminiRequest.stream) { + return await geminiHandleStreamGenerateContent(req, res) + } else { + return await geminiHandleGenerateContent(req, res) + } + } else { + return res.status(500).json({ + error: { + message: `Unsupported backend: ${backend}`, + type: 'server_error', + code: 'unsupported_backend' + } + }) + } +} + +// 🔄 OpenAI 兼容的 chat/completions 端点（智能后端路由） +router.post('/v1/chat/completions', authenticateApiKey, async (req, res) => { + try { + // 验证必需参数 + if (!req.body.messages || !Array.isArray(req.body.messages) || req.body.messages.length === 0) { + return res.status(400).json({ + error: { + message: 'Messages array is required and cannot be empty', + type: 'invalid_request_error', + code: 'invalid_request' + } + }) + } + + const requestedModel = req.body.model || 'claude-3-5-sonnet-20241022' + req.body.model = requestedModel // 确保模型已设置 + + // 使用统一的后端路由处理器 + await routeToBackend(req, res, requestedModel) + } catch (error) { + logger.error('❌ OpenAI chat/completions error:', error) + if (!res.headersSent) { + res.status(500).json({ + error: { + message: 'Internal server error', + type: 'server_error', + code: 'internal_error' + } + }) + } + } +}) + +// 🔄 OpenAI 兼容的 completions 端点（传统格式，智能后端路由） +router.post('/v1/completions', authenticateApiKey, async (req, res) => { + try { + // 验证必需参数 + if (!req.body.prompt) { + return res.status(400).json({ + error: { + message: 'Prompt is required', + type: 'invalid_request_error', + code: 'invalid_request' + } + }) + } + + // 将传统 completions 格式转换为 chat 格式 + const originalBody = req.body + const requestedModel = originalBody.model || 'claude-3-5-sonnet-20241022' + + req.body = { + model: requestedModel, + messages: [ + { + role: 'user', + content: originalBody.prompt + } + ], + max_tokens: originalBody.max_tokens, + temperature: originalBody.temperature, + top_p: originalBody.top_p, + stream: originalBody.stream, + stop: originalBody.stop, + n: originalBody.n || 1, + presence_penalty: originalBody.presence_penalty, + frequency_penalty: originalBody.frequency_penalty, + logit_bias: originalBody.logit_bias, + user: originalBody.user + } + + // 使用统一的后端路由处理器 + await routeToBackend(req, res, requestedModel) + } catch (error) { + logger.error('❌ OpenAI completions error:', error) + if (!res.headersSent) { + res.status(500).json({ + error: { + message: 'Failed to process completion request', + type: 'server_error', + code: 'internal_error' + } + }) + } + } +}) + +module.exports = router +module.exports.detectBackendFromModel = detectBackendFromModel +module.exports.routeToBackend = routeToBackend From 6bb74376caecaff0c3c6c59326667989e07fde25 Mon Sep 17 00:00:00 2001 From: jft0m <377632523@qq.com> Date: Tue, 14 Oct 2025 16:17:12 +0000 Subject: [PATCH 3/3] fix: add /openai/v1/chat/completions route support MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Register unifiedRoutes under /openai prefix to enable /openai/v1/chat/completions - Reuse existing intelligent routing logic from unified.js (no code duplication) - Keep existing Codex API routes (/openai/responses, /openai/v1/responses) unchanged Benefits: - Fixes 404 error for /openai/v1/chat/completions endpoint - Provides consistent API experience across /api and /openai prefixes - Automatically routes to correct backend (Claude/OpenAI/Gemini) based on model Tested: - ✅ /openai/v1/chat/completions now returns authentication error (route works) - ✅ /api/v1/chat/completions continues to work - ✅ Existing Codex routes remain functional 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude --- src/app.js | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/app.js b/src/app.js index 4f86c6cf..67454ea8 100644 --- a/src/app.js +++ b/src/app.js @@ -269,7 +269,8 @@ class Application { this.app.use('/gemini', geminiRoutes) // 保留原有路径以保持向后兼容 this.app.use('/openai/gemini', openaiGeminiRoutes) this.app.use('/openai/claude', openaiClaudeRoutes) - this.app.use('/openai', openaiRoutes) + this.app.use('/openai', unifiedRoutes) // 复用统一智能路由，支持 /openai/v1/chat/completions + this.app.use('/openai', openaiRoutes) // Codex API 路由（/openai/responses, /openai/v1/responses） // Droid 路由：支持多种 Factory.ai 端点 this.app.use('/droid', droidRoutes) // Droid (Factory.ai) API 转发 this.app.use('/azure', azureOpenaiRoutes)