Upload 5 files

Calmlo OpenAI Proxy Dockerfile

@@ -0,0 +1,24 @@
+# 使用官方 Node.js 18 LTS 镜像作为基础
+FROM node:18-alpine
+
+# 设置工作目录
+WORKDIR /usr/src/app
+
+# 复制 package.json 和 package-lock.json (如果存在)
+COPY package*.json ./
+
+# 安装项目依赖
+RUN npm install
+
+# 复制应用源代码
+COPY . .
+
+# 暴露应用程序使用的端口
+EXPOSE 3000
+
+# 定义环境变量 (可以在 docker-compose 中覆盖)
+ENV PORT=3000
+# FAL_KEY 应该在运行时通过 docker-compose 传入，而不是硬编码在这里
+
+# 运行应用程序的命令
+CMD [ "npm", "start" ]

Calmlo fal-openai-proxy README.md

@@ -0,0 +1,30 @@
+---
+title: fal-openai-proxy
+emoji: 🐳
+colorFrom: red
+colorTo: red
+sdk: docker
+sdk_version: "{{sdkVersion}}"
+app_file: app.py
+pinned: false
+app_port: 3000
+---
+
+
+
+## openai请求格式转fal
+输入限制： System prompt 和 prompt 分别最长为 5000 字符（不是 token）。
+
+输出长度： 测试了下输出长篇小说，出了 5W 多 token。
+
+上下文： 不支持。
+
+于是用 gemini 糊了个 openaiToFal 的服务，模拟上下文以 5000 字符为分界线，分别塞到 System prompt 和 prompt，这样可以把输入扩展到 1W 字符，太早的聊天记录会被顶掉。github 地址是一个 docker compose 包，把你的 key 填入 docker-compose.yml，一键启动 docker compose up -d 即可。默认端口 13000。
+
+## 部署步骤
+1、修改docker-compose.yml填入fal的api key
+
+2、`docker compose up -d`启动
+
+## 重要
+我是搭配newapi管理使用，所以**没有鉴权**，有需要自己加。

fal-openai-proxy docker-compose.yml

@@ -0,0 +1,15 @@
+services:
+  fal-openai-proxy:
+    build: . # 构建当前目录下的 Dockerfile
+    container_name: fal_openai_proxy
+    ports:
+      - "13000:3000" # 将主机的 3000 端口映射到容器的 3000 端口
+    environment:
+      # 在这里设置你的 Fal AI API Key
+      # 或者，为了安全起见，你可以创建一个 .env 文件，并在其中定义 FAL_KEY
+      # 然后取消下面行的注释:
+      # env_file:
+      #  - .env
+      FAL_KEY: "" # !! 重要：替换为你的真实 Fal AI Key !!
+      PORT: 3000 # 确保容器内的端口与 Dockerfile 和 server.js 中一致
+    restart: unless-stopped # 服务失败时自动重启，除非手动停止

fal-openai-proxy package.json

@@ -0,0 +1,17 @@
+{
+  "name": "fal-openai-proxy",
+  "version": "1.0.0",
+  "description": "Proxy server to convert OpenAI requests to fal-ai format",
+  "main": "server.js",
+  "type": "module",
+  "scripts": {
+    "start": "node server.js"
+  },
+  "dependencies": {
+    "@fal-ai/client": "latest",
+    "express": "^4.19.2"
+  },
+  "engines": {
+    "node": ">=18.0.0"
+  }
+}

fal-openai-proxy server.js

@@ -0,0 +1,550 @@
+import express from 'express';
+import { fal } from '@fal-ai/client';
+
+// --- Multi-Key Configuration ---
+// *** 使用 FAL_KEY 环境变量读取逗号分隔的密钥 ***
+const rawFalKeys = process.env.FAL_KEY; // Expect comma-separated keys: key1,key2,key3 in FAL_KEY
+const API_KEY = process.env.API_KEY; // Custom API Key for proxy auth remains the same
+
+if (!rawFalKeys) {
+    // *** 更新错误信息以引用 FAL_KEY ***
+    console.error("Error: FAL_KEY environment variable is not set (should be comma-separated).");
+    process.exit(1);
+}
+
+if (!API_KEY) {
+    console.error("Error: API_KEY environment variable is not set.");
+    process.exit(1);
+}
+
+// Parse and prepare the keys
+let falKeys = rawFalKeys.split(',')
+    .map(key => key.trim())
+    .filter(key => key.length > 0)
+    .map(key => ({
+        key: key,
+        failed: false,          // Track if the key is currently considered failed
+        failedTimestamp: 0      // Timestamp when the key was marked as failed
+    }));
+
+if (falKeys.length === 0) {
+    // *** 更新错误信息以引用 FAL_KEY ***
+    console.error("Error: No valid keys found in FAL_KEY after processing the environment variable.");
+    process.exit(1);
+}
+
+let currentKeyIndex = 0;
+const failedKeyCooldown = 60 * 1000; // Cooldown period in milliseconds (e.g., 60 seconds) before retrying a failed key
+
+// *** 更新日志信息以引用 FAL_KEY ***
+console.log(`Loaded ${falKeys.length} FAL API Key(s) from FAL_KEY environment variable.`);
+console.log(`Failed key cooldown period: ${failedKeyCooldown / 1000} seconds.`);
+
+// NOTE: We will configure fal client per request now, so initial global config is removed.
+
+// --- Key Management Functions ---
+
+/**
+ * Selects the next available FAL key using round-robin and skipping recently failed keys.
+ * @returns {object | null} Key info object { key, failed, failedTimestamp } or null if all keys are failed.
+ */
+function getNextKey() {
+    const totalKeys = falKeys.length;
+    if (totalKeys === 0) return null;
+
+    let attempts = 0;
+    while (attempts < totalKeys) {
+        const keyIndex = currentKeyIndex % totalKeys;
+        const keyInfo = falKeys[keyIndex];
+        // Increment index for the *next* call, ensuring round-robin
+        currentKeyIndex = (currentKeyIndex + 1) % totalKeys;
+
+        // Check if key is marked as failed and if cooldown has passed
+        if (keyInfo.failed) {
+            const now = Date.now();
+            if (now - keyInfo.failedTimestamp < failedKeyCooldown) {
+                // console.log(`Key index ${keyIndex} is in cooldown. Skipping.`);
+                attempts++;
+                continue; // Skip this key, it's still in cooldown
+            } else {
+                console.log(`Cooldown finished for key index ${keyIndex}. Resetting failure status.`);
+                keyInfo.failed = false; // Cooldown expired, reset status
+                keyInfo.failedTimestamp = 0;
+            }
+        }
+        // console.log(`Selected key index: ${keyIndex}`);
+        return keyInfo; // Return the valid key info object
+    }
+
+    console.warn("All FAL keys are currently marked as failed and in cooldown.");
+    return null; // All keys are currently failed and within cooldown
+}
+
+/**
+ * Marks a specific key as failed.
+ * @param {object} keyInfo - The key info object to mark as failed.
+ */
+function markKeyFailed(keyInfo) {
+    if (keyInfo && !keyInfo.failed) { // Only mark if not already marked
+        keyInfo.failed = true;
+        keyInfo.failedTimestamp = Date.now();
+        const keyIndex = falKeys.findIndex(k => k.key === keyInfo.key);
+        console.warn(`Marking key index ${keyIndex} (ending ...${keyInfo.key.slice(-4)}) as failed.`);
+    }
+}
+
+/**
+ * Determines if an error likely indicates an API key issue (auth, quota, etc.).
+ * This needs refinement based on actual errors from fal.ai.
+ * @param {Error} error - The error object caught from the fal client.
+ * @returns {boolean} - True if the error suggests a key failure, false otherwise.
+ */
+function isKeyRelatedError(error) {
+    const errorMessage = error?.message?.toLowerCase() || '';
+    const errorStatus = error?.status; // Assuming the error object might have a status property
+
+    // Check for common indicators of key issues
+    if (errorStatus === 401 || errorStatus === 403 || // Unauthorized, Forbidden
+        errorMessage.includes('authentication failed') ||
+        errorMessage.includes('invalid api key') ||
+        errorMessage.includes('permission denied')) {
+        return true;
+    }
+    if (errorStatus === 429 || // Too Many Requests (Rate Limit / Quota)
+        errorMessage.includes('rate limit exceeded') ||
+        errorMessage.includes('quota exceeded')) {
+        return true;
+    }
+    // Add more specific error messages or codes from fal.ai if known
+    // console.log("Error does not appear to be key-related:", error); // Debugging
+    return false;
+}
+
+// --- Express App Setup ---
+const app = express();
+app.use(express.json({ limit: '50mb' }));
+app.use(express.urlencoded({ extended: true, limit: '50mb' }));
+
+const PORT = process.env.PORT || 3000;
+
+// API Key 鉴权中间件 (Remains the same, checks custom API_KEY)
+const apiKeyAuth = (req, res, next) => {
+    const authHeader = req.headers['authorization'];
+
+    if (!authHeader) {
+        console.warn('Unauthorized: No Authorization header provided');
+        return res.status(401).json({ error: 'Unauthorized: No API Key provided' });
+    }
+
+    const authParts = authHeader.split(' ');
+    if (authParts.length !== 2 || authParts[0].toLowerCase() !== 'bearer') {
+        console.warn('Unauthorized: Invalid Authorization header format');
+        return res.status(401).json({ error: 'Unauthorized: Invalid Authorization header format' });
+    }
+
+    const providedKey = authParts[1];
+    if (providedKey !== API_KEY) {
+        console.warn('Unauthorized: Invalid API Key');
+        return res.status(401).json({ error: 'Unauthorized: Invalid API Key' });
+    }
+
+    next();
+};
+
+app.use(['/v1/models', '/v1/chat/completions'], apiKeyAuth);
+
+// === 全局定义限制 === (Remains the same)
+const PROMPT_LIMIT = 4800;
+const SYSTEM_PROMPT_LIMIT = 4800;
+// === 限制定义结束 ===
+
+// 定义 fal-ai/any-llm 支持的模型列表 (Remains the same)
+const FAL_SUPPORTED_MODELS = [
+    "anthropic/claude-3.7-sonnet",
+    "anthropic/claude-3.5-sonnet",
+    "anthropic/claude-3-5-haiku",
+    "anthropic/claude-3-haiku",
+    "google/gemini-pro-1.5",
+    "google/gemini-flash-1.5",
+    "google/gemini-flash-1.5-8b",
+    "google/gemini-2.0-flash-001",
+    "meta-llama/llama-3.2-1b-instruct",
+    "meta-llama/llama-3.2-3b-instruct",
+    "meta-llama/llama-3.1-8b-instruct",
+    "meta-llama/llama-3.1-70b-instruct",
+    "openai/gpt-4o-mini",
+    "openai/gpt-4o",
+    "deepseek/deepseek-r1",
+    "meta-llama/llama-4-maverick",
+    "meta-llama/llama-4-scout"
+];
+
+// Helper function to get owner from model ID (Remains the same)
+const getOwner = (modelId) => {
+    if (modelId && modelId.includes('/')) {
+        return modelId.split('/')[0];
+    }
+    return 'fal-ai';
+};
+
+// GET /v1/models endpoint (Remains the same)
+app.get('/v1/models', (req, res) => {
+    console.log("Received request for GET /v1/models");
+    try {
+        const modelsData = FAL_SUPPORTED_MODELS.map(modelId => ({
+            id: modelId, object: "model", created: 1700000000, owned_by: getOwner(modelId)
+        }));
+        res.json({ object: "list", data: modelsData });
+        console.log("Successfully returned model list.");
+    } catch (error) {
+        console.error("Error processing GET /v1/models:", error);
+        res.status(500).json({ error: "Failed to retrieve model list." });
+    }
+});
+
+// === convertMessagesToFalPrompt 函数 (Remains the same) ===
+function convertMessagesToFalPrompt(messages) {
+    let fixed_system_prompt_content = "";
+    const conversation_message_blocks = [];
+    // console.log(`Original messages count: ${messages.length}`); // Less verbose logging
+
+    // 1. 分离 System 消息，格式化 User/Assistant 消息
+    for (const message of messages) {
+        let content = (message.content === null || message.content === undefined) ? "" : String(message.content);
+        switch (message.role) {
+            case 'system':
+                fixed_system_prompt_content += `System: ${content}\n\n`;
+                break;
+            case 'user':
+                conversation_message_blocks.push(`Human: ${content}\n\n`);
+                break;
+            case 'assistant':
+                conversation_message_blocks.push(`Assistant: ${content}\n\n`);
+                break;
+            default:
+                console.warn(`Unsupported role: ${message.role}`);
+                continue;
+        }
+    }
+
+    // 2. 截断合并后的 system 消息（如果超长）
+    if (fixed_system_prompt_content.length > SYSTEM_PROMPT_LIMIT) {
+        const originalLength = fixed_system_prompt_content.length;
+        fixed_system_prompt_content = fixed_system_prompt_content.substring(0, SYSTEM_PROMPT_LIMIT);
+        console.warn(`Combined system messages truncated from ${originalLength} to ${SYSTEM_PROMPT_LIMIT}`);
+    }
+    fixed_system_prompt_content = fixed_system_prompt_content.trim();
+
+
+    // 3. 计算 system_prompt 中留给对话历史的剩余空间
+    let space_occupied_by_fixed_system = 0;
+    if (fixed_system_prompt_content.length > 0) {
+         space_occupied_by_fixed_system = fixed_system_prompt_content.length + 4; // 预留 \n\n...\n\n 的长度
+    }
+     const remaining_system_limit = Math.max(0, SYSTEM_PROMPT_LIMIT - space_occupied_by_fixed_system);
+    // console.log(`Trimmed fixed system prompt length: ${fixed_system_prompt_content.length}. Approx remaining system history limit: ${remaining_system_limit}`);
+
+
+    // 4. 反向填充 User/Assistant 对话历史
+    const prompt_history_blocks = [];
+    const system_prompt_history_blocks = [];
+    let current_prompt_length = 0;
+    let current_system_history_length = 0;
+    let promptFull = false;
+    let systemHistoryFull = (remaining_system_limit <= 0);
+
+    // console.log(`Processing ${conversation_message_blocks.length} user/assistant messages for recency filling.`);
+    for (let i = conversation_message_blocks.length - 1; i >= 0; i--) {
+        const message_block = conversation_message_blocks[i];
+        const block_length = message_block.length;
+
+        if (promptFull && systemHistoryFull) {
+            // console.log(`Both prompt and system history slots full. Omitting older messages from index ${i}.`);
+            break;
+        }
+
+        // 优先尝试放入 prompt
+        if (!promptFull) {
+            if (current_prompt_length + block_length <= PROMPT_LIMIT) {
+                prompt_history_blocks.unshift(message_block);
+                current_prompt_length += block_length;
+                continue;
+            } else {
+                promptFull = true;
+                // console.log(`Prompt limit (${PROMPT_LIMIT}) reached. Trying system history slot.`);
+            }
+        }
+
+        // 如果 prompt 满了，尝试放入 system_prompt 的剩余空间
+        if (!systemHistoryFull) {
+            if (current_system_history_length + block_length <= remaining_system_limit) {
+                 system_prompt_history_blocks.unshift(message_block);
+                 current_system_history_length += block_length;
+                 continue;
+            } else {
+                 systemHistoryFull = true;
+                 // console.log(`System history limit (${remaining_system_limit}) reached.`);
+            }
+        }
+    }
+
+    // 5. *** 组合最终的 prompt 和 system_prompt (包含分隔符逻辑) ***
+    const system_prompt_history_content = system_prompt_history_blocks.join('').trim();
+    const final_prompt = prompt_history_blocks.join('').trim();
+
+    // 定义分隔符
+    const SEPARATOR = "\n\n-------下面是比较早之前的对话内容-----\n\n";
+
+    let final_system_prompt = "";
+
+    const hasFixedSystem = fixed_system_prompt_content.length > 0;
+    const hasSystemHistory = system_prompt_history_content.length > 0;
+
+    if (hasFixedSystem && hasSystemHistory) {
+        final_system_prompt = fixed_system_prompt_content + SEPARATOR + system_prompt_history_content;
+        // console.log("Combining fixed system prompt and history with separator.");
+    } else if (hasFixedSystem) {
+        final_system_prompt = fixed_system_prompt_content;
+        // console.log("Using only fixed system prompt.");
+    } else if (hasSystemHistory) {
+        final_system_prompt = system_prompt_history_content;
+        // console.log("Using only history in system prompt slot.");
+    }
+
+    // 6. 返回结果
+    const result = {
+        system_prompt: final_system_prompt,
+        prompt: final_prompt
+    };
+
+    console.log(`Final system_prompt length: ${result.system_prompt.length}, Final prompt length: ${result.prompt.length}`);
+
+    return result;
+}
+// === convertMessagesToFalPrompt 函数结束 ===
+
+
+/**
+ * Wraps the fal.ai API call with retry logic using available keys.
+ * @param {'stream' | 'subscribe'} operation - The fal operation to perform.
+ * @param {string} functionId - The fal function ID (e.g., "fal-ai/any-llm").
+ * @param {object} params - The parameters for the fal function call (input, logs, etc.).
+ * @returns {Promise<any>} - The result from the successful fal call (stream or subscription result).
+ * @throws {Error} - Throws an error if all keys fail or a non-key-related error occurs.
+ */
+async function tryFalCallWithFailover(operation, functionId, params) {
+    const maxRetries = falKeys.length; // Try each key at most once per request cycle
+    let lastError = null;
+
+    for (let i = 0; i < maxRetries; i++) {
+        const keyInfo = getNextKey();
+        if (!keyInfo) {
+            throw new Error(lastError ? `All FAL keys failed. Last error: ${lastError.message}` : "All FAL keys are currently unavailable (failed or in cooldown).");
+        }
+
+        const currentFalKey = keyInfo.key;
+        console.log(`Attempt ${i + 1}/${maxRetries}: Using key ending in ...${currentFalKey.slice(-4)}`);
+
+        try {
+            // --- Configure fal client with the selected key for this attempt ---
+            // WARNING: This global config change might have concurrency issues in high-load scenarios
+            // if the fal client library doesn't isolate requests properly.
+            fal.config({ credentials: currentFalKey });
+
+            if (operation === 'stream') {
+                const streamResult = await fal.stream(functionId, params);
+                console.log(`Successfully initiated stream with key ending in ...${currentFalKey.slice(-4)}`);
+                return streamResult;
+            } else { // 'subscribe' (non-stream)
+                const result = await fal.subscribe(functionId, params);
+                console.log(`Successfully completed subscribe request with key ending in ...${currentFalKey.slice(-4)}`);
+
+                if (result && result.error) {
+                     console.warn(`Fal-ai returned an application error (non-stream) with key ...${currentFalKey.slice(-4)}: ${JSON.stringify(result.error)}`);
+                }
+                return result;
+            }
+        } catch (error) {
+            console.error(`Error using key ending in ...${currentFalKey.slice(-4)}:`, error.message || error);
+            lastError = error;
+
+            if (isKeyRelatedError(error)) {
+                markKeyFailed(keyInfo);
+                console.log(`Key marked as failed. Trying next key if available...`);
+            } else {
+                console.error("Non-key related error occurred. Aborting retries.");
+                throw error;
+            }
+        }
+    }
+
+    console.error("All FAL keys failed after attempting each one.");
+    throw new Error(lastError ? `All FAL keys failed. Last error: ${lastError.message}` : "All FAL API keys failed.");
+}
+
+
+// POST /v1/chat/completions endpoint (Modified to use tryFalCallWithFailover)
+app.post('/v1/chat/completions', async (req, res) => {
+    const { model, messages, stream = false, reasoning = false, ...restOpenAIParams } = req.body;
+
+    console.log(`Received chat completion request for model: ${model}, stream: ${stream}`);
+
+    if (!FAL_SUPPORTED_MODELS.includes(model)) {
+         console.warn(`Warning: Requested model '${model}' is not in the explicitly supported list.`);
+    }
+    if (!model || !messages || !Array.isArray(messages) || messages.length === 0) {
+        console.error("Invalid request parameters:", { model, messages: Array.isArray(messages) ? messages.length : typeof messages });
+        return res.status(400).json({ error: 'Missing or invalid parameters: model and messages array are required.' });
+    }
+
+    try {
+        const { prompt, system_prompt } = convertMessagesToFalPrompt(messages);
+
+        const falInput = {
+            model: model,
+            prompt: prompt,
+            ...(system_prompt && { system_prompt: system_prompt }),
+            reasoning: !!reasoning,
+        };
+
+        console.log("Prepared Fal Input (lengths):", { system_prompt: system_prompt?.length, prompt: prompt?.length });
+
+        if (stream) {
+            res.setHeader('Content-Type', 'text/event-stream; charset=utf-8');
+            res.setHeader('Cache-Control', 'no-cache');
+            res.setHeader('Connection', 'keep-alive');
+            res.setHeader('Access-Control-Allow-Origin', '*');
+            res.flushHeaders();
+
+            let previousOutput = '';
+            let falStream;
+
+            try {
+                 falStream = await tryFalCallWithFailover('stream', "fal-ai/any-llm", { input: falInput });
+
+                 for await (const event of falStream) {
+                    const currentOutput = (event && typeof event.output === 'string') ? event.output : '';
+                    const isPartial = (event && typeof event.partial === 'boolean') ? event.partial : true;
+                    const errorInfo = (event && event.error) ? event.error : null;
+
+                    if (errorInfo) {
+                        console.error("Error received *during* fal stream:", errorInfo);
+                        const errorChunk = { id: `chatcmpl-${Date.now()}-error`, object: "chat.completion.chunk", created: Math.floor(Date.now() / 1000), model: model, choices: [{ index: 0, delta: {}, finish_reason: "error", message: { role: 'assistant', content: `Fal Stream Error: ${JSON.stringify(errorInfo)}` } }] };
+                        res.write(`data: ${JSON.stringify(errorChunk)}\n\n`);
+                        break;
+                    }
+
+                    let deltaContent = '';
+                    if (currentOutput.startsWith(previousOutput)) {
+                        deltaContent = currentOutput.substring(previousOutput.length);
+                    } else if (currentOutput.length > 0) {
+                        console.warn("Fal stream output mismatch detected. Sending full current output as delta.", { previousLength: previousOutput.length, currentLength: currentOutput.length });
+                        deltaContent = currentOutput;
+                        previousOutput = '';
+                    }
+                    previousOutput = currentOutput;
+
+                    if (deltaContent || !isPartial) {
+                        const openAIChunk = { id: `chatcmpl-${Date.now()}`, object: "chat.completion.chunk", created: Math.floor(Date.now() / 1000), model: model, choices: [{ index: 0, delta: { content: deltaContent }, finish_reason: isPartial === false ? "stop" : null }] };
+                        res.write(`data: ${JSON.stringify(openAIChunk)}\n\n`);
+                    }
+                 }
+                 res.write(`data: [DONE]\n\n`);
+                 res.end();
+                 console.log("Stream finished successfully.");
+
+            } catch (streamError) {
+                console.error('Error during stream processing:', streamError);
+                if (!res.writableEnded) {
+                     try {
+                         const errorDetails = (streamError instanceof Error) ? streamError.message : JSON.stringify(streamError);
+                         const finalErrorChunk = { error: { message: "Stream failed", type: "proxy_error", details: errorDetails } };
+                         res.write(`data: ${JSON.stringify(finalErrorChunk)}\n\n`);
+                         res.write(`data: [DONE]\n\n`);
+                         res.end();
+                     } catch (finalError) {
+                         console.error('Error sending final stream error message to client:', finalError);
+                         if (!res.writableEnded) { res.end(); }
+                     }
+                 }
+            }
+
+        } else { // Non-stream
+             console.log("Executing non-stream request with failover...");
+             const result = await tryFalCallWithFailover('subscribe', "fal-ai/any-llm", { input: falInput, logs: true });
+
+             console.log("Received non-stream result from fal-ai via failover wrapper.");
+
+             if (result && result.error) {
+                 console.error("Fal-ai returned an application error in non-stream mode (after successful API call):", result.error);
+                 return res.status(500).json({
+                      object: "error",
+                      message: `Fal-ai application error: ${JSON.stringify(result.error)}`,
+                      type: "fal_ai_error",
+                      param: null,
+                      code: result.error.code || null
+                 });
+             }
+
+             const openAIResponse = {
+                 id: `chatcmpl-${result?.requestId || Date.now()}`,
+                 object: "chat.completion",
+                 created: Math.floor(Date.now() / 1000),
+                 model: model,
+                 choices: [{
+                     index: 0,
+                     message: {
+                         role: "assistant",
+                         content: result?.output || ""
+                     },
+                     finish_reason: "stop"
+                 }],
+                 usage: {
+                     prompt_tokens: null,
+                     completion_tokens: null,
+                     total_tokens: null
+                 },
+                 system_fingerprint: null,
+                 ...(result?.reasoning && { fal_reasoning: result.reasoning }),
+             };
+             res.json(openAIResponse);
+             console.log("Returned non-stream response successfully.");
+        }
+
+    } catch (error) {
+        console.error('Unhandled error in /v1/chat/completions:', error);
+        if (!res.headersSent) {
+            const errorMessage = (error instanceof Error) ? error.message : JSON.stringify(error);
+            const errorType = error.message?.includes("All FAL keys failed") ? "api_key_error" : "proxy_internal_error";
+            res.status(500).json({
+                 error: {
+                     message: `Internal Server Error in Proxy: ${errorMessage}`,
+                     type: errorType,
+                     details: error.stack // Optional: include stack in dev/debug mode
+                 }
+            });
+        } else if (!res.writableEnded) {
+             console.error("Headers already sent, attempting to end response after error.");
+             res.end();
+        }
+    }
+});
+
+// --- Server Start ---
+app.listen(PORT, () => {
+    console.log(`===========================================================`);
+    console.log(` Fal OpenAI Proxy Server (Multi-Key Failover)`);
+    console.log(` Listening on port: ${PORT}`);
+    // *** 更新日志信息以引用 FAL_KEY ***
+    console.log(` Loaded ${falKeys.length} FAL API Key(s) from FAL_KEY.`);
+    console.log(` API Key Auth Enabled: ${API_KEY ? 'Yes' : 'No'}`);
+    console.log(` Limits: System Prompt=${SYSTEM_PROMPT_LIMIT}, Prompt=${PROMPT_LIMIT}`);
+    console.log(` Chat Completions: POST http://localhost:${PORT}/v1/chat/completions`);
+    console.log(` Models Endpoint: GET http://localhost:${PORT}/v1/models`);
+    console.log(`===========================================================`);
+});
+
+// Root path response
+app.get('/', (req, res) => {
+    res.send('Fal OpenAI Proxy (Multi-Key Failover) is running.');
+});