app.js

import express from 'express';
import cors from 'cors';
import dotenv from 'dotenv';
import OpenAI from "openai";
import { BedrockRuntimeClient, ConverseCommand, ConverseStreamCommand } from "@aws-sdk/client-bedrock-runtime";
import { NodeHttpHandler } from "@smithy/node-http-handler";

dotenv.config();
const app = express();
const PORT = process.env.PORT || 7860;

app.use(cors());
app.use(express.json({ limit: '50mb' })); 

// --- SYSTEM PROMPT DEFINITIONS ---
const CLAUDE_SYSTEM_PROMPT = "You are a pro. Provide elite, high-level technical responses.";
const GPT_SYSTEM_PROMPT = "You are a worker. Be concise, efficient, and get the job done.";

const bedrockClient = new BedrockRuntimeClient({
    region: "us-east-1",
    requestHandler: new NodeHttpHandler({
        http2Handler: undefined, 
    }) 
});

const azureOpenAI = new OpenAI({
    apiKey: "",
    baseURL: ``,
    defaultQuery: { "api-version": "2024-05-01-preview" },
    defaultHeaders: { "api-key": "" }
});

// --- DYNAMIC MODEL ROUTER ---
function getBedrockModelId(modelName) {
    switch(modelName) {
        case "haiku":
            return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/global.anthropic.claude-haiku-4-5-20251001-v1:0";
        case "maverick":
            return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/us.meta.llama4-maverick-17b-instruct-v1:0";
        case "claude":
        default:
            return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/global.anthropic.claude-sonnet-4-6";
    }
}

// --- NON-STREAMING ENDPOINT (UPDATED FOR VISION) ---
app.post('/api/generate', async (req, res) => {
    // EXTRACT IMAGES HERE
    const { model, prompt, system_prompt, images } = req.body;
    console.log(`[TRAFFIC] Request for ${model} ${images?.length ? 'with images' : ''}`);

    try {
        if (model === "gpt" || model === "gpt-5-mini") {
            let messagesPayload =[
                { role: "system", content: system_prompt || GPT_SYSTEM_PROMPT }
            ];

            // VISION SUPPORT FOR AZURE OPENAI
            if (images && images.length > 0) {
                let userContent = [{ type: "text", text: prompt }];
                images.forEach(imgStr => {
                    userContent.push({ type: "image_url", image_url: { url: imgStr } });
                });
                messagesPayload.push({ role: "user", content: userContent });
            } else {
                messagesPayload.push({ role: "user", content: prompt });
            }

            const response = await azureOpenAI.chat.completions.create({
                model: "gpt-5-mini",
                messages: messagesPayload,
                reasoning_effort: "high"
            });
            
            const totalTokens = response.usage ? response.usage.total_tokens : 0;
            res.json({ success: true, data: response.choices[0].message.content, usage: { totalTokenCount: totalTokens } });

        } else {
            // Handles Claude Sonnet, Claude Haiku, and Llama Maverick
            const bedrockModelId = getBedrockModelId(model);
            
            // VISION SUPPORT FOR AWS BEDROCK
            let contentBlock = [{ text: prompt }];
            if (images && images.length > 0) {
                const imageBlocks = images.map(imgStr => {
                    const base64Data = imgStr.replace(/^data:image\/\w+;base64,/, "");
                    return {
                        image: {
                            format: 'png', // Assuming normalized to PNG by frontend
                            source: { bytes: Buffer.from(base64Data, 'base64') }
                        }
                    };
                });
                contentBlock = [...imageBlocks, ...contentBlock];
            }

            const command = new ConverseCommand({
                modelId: bedrockModelId,
                system:[{ text: system_prompt || CLAUDE_SYSTEM_PROMPT }],
                messages: [{ role: "user", content: contentBlock }],
                
                // Ensure maxTokens is large enough for reasoning + response
                inferenceConfig: { 
                    maxTokens: model.includes("haiku") ? 32000 : 4000, 
                    temperature: 1 
                },

                performanceConfig: model.includes("maverick") ? { latency: "standard" } : undefined,

                additionalModelRequestFields: (function() {
                    if (model.includes("haiku")) {
                        return {
                            reasoning_config: {
                                type: "enabled",
                                budget_tokens: 2048 
                            } 
                        };
                    } else if (model.includes("claude")) {
                        return {
                            // thinking: { type: "adaptive" },
                            output_config: { effort: "high" }
                        };
                    }
                    return undefined;
                })()
            });
          
            const response = await bedrockClient.send(command);
  
          const text = response.output.message.content.find(b => b.text)?.text;
            const tokenUsage = response.usage ? (response.usage.inputTokens + response.usage.outputTokens) : 0;

            res.json({ success: true, data: text, usage: { totalTokenCount: tokenUsage } });
        }
    } catch (err) {
        console.error(`❌[${model?.toUpperCase() || 'UNKNOWN'} ERROR]:`, err.name, err.message);
        res.status(500).json({ success: false, error: `${err.name}: ${err.message}` });
    }
});

// --- STREAMING ENDPOINT ---
app.post('/api/stream', async (req, res) => {
    const { model, prompt, system_prompt, images } = req.body; 
    console.log(`[STREAM] Request for ${model} ${images?.length ? 'with images' : ''}`);

    res.setHeader('Content-Type', 'text/plain; charset=utf-8');
    res.setHeader('Transfer-Encoding', 'chunked');
    res.setHeader('X-Accel-Buffering', 'no'); 
    res.flushHeaders();

    let totalTokenCount = 0;

    try {
        if (model === "gpt" || model === "gpt-5-mini") {
            let messagesPayload =[
                { role: "system", content: system_prompt || GPT_SYSTEM_PROMPT }
            ];

            let userContent =[];
            if (images && images.length > 0) {
                userContent.push({ type: "text", text: prompt });
                images.forEach(imgStr => {
                    userContent.push({ type: "image_url", image_url: { url: imgStr } });
                });
                messagesPayload.push({ role: "user", content: userContent });
            } else {
                messagesPayload.push({ role: "user", content: prompt });
            }

            const stream = await azureOpenAI.chat.completions.create({
                model: "gpt-5-mini",
                messages: messagesPayload,
                reasoning_effort: "high",
                stream: true,
                stream_options: { include_usage: true } 
            });

            for await (const chunk of stream) {
                const delta = chunk.choices[0]?.delta;
                if (delta?.reasoning_content) res.write(`__THINK__${delta.reasoning_content}`);
                else if (delta?.content) res.write(delta.content);
                if (chunk.usage) totalTokenCount = chunk.usage.total_tokens;
            }
            
            res.write(`__USAGE__${JSON.stringify({ totalTokenCount })}`);
            res.end();

        } else {
            const bedrockModelId = getBedrockModelId(model);
            let contentBlock = [{ text: prompt }];
            
            if (images && images.length > 0) {
                const imageBlocks = images.map(imgStr => {
                    const base64Data = imgStr.replace(/^data:image\/\w+;base64,/, "");
                    return {
                        image: {
                            format: 'png', // Assuming normalized to PNG by frontend
                            source: { bytes: Buffer.from(base64Data, 'base64') }
                        }
                    };
                });
                contentBlock = [...imageBlocks, ...contentBlock];
            }

            const command = new ConverseStreamCommand({
                modelId: bedrockModelId,
                system:[{ text: system_prompt || CLAUDE_SYSTEM_PROMPT }],
                messages:[{ role: "user", content: contentBlock }],
                inferenceConfig: { maxTokens: 48000, temperature: 1 },
                additionalModelRequestFields: model.includes("claude") ? { 
                    thinking: { type: "adaptive" },
                    output_config: { effort: "high" }
                } : undefined
            });

            const response = await bedrockClient.send(command);
            
            for await (const chunk of response.stream) {
                if (chunk.contentBlockDelta) {
                    const delta = chunk.contentBlockDelta.delta;
                    if (delta.reasoningContent && delta.reasoningContent.text) {
                        res.write(`__THINK__${delta.reasoningContent.text}`);
                    } else if (delta.text) {
                        res.write(delta.text);
                    }
                }
                if (chunk.metadata && chunk.metadata.usage) {
                    totalTokenCount = (chunk.metadata.usage.inputTokens || 0) + (chunk.metadata.usage.outputTokens || 0);
                }
            }
            
            res.write(`__USAGE__${JSON.stringify({ totalTokenCount })}`);
            res.end();
        }
    } catch (err) {
        console.error(`❌ [STREAM ERROR]:`, err.message);
        res.write(`ERROR: ${err.message}`);
        res.end();
    }
});

app.get('/', async (req, res) => { res.json({ success: true }); });
app.listen(PORT, '0.0.0.0', () => console.log(`Main AI Agent live on port ${PORT}`));


/* import express from 'express';
import cors from 'cors';
import dotenv from 'dotenv';
import OpenAI from "openai";
import { BedrockRuntimeClient, ConverseCommand, ConverseStreamCommand } from "@aws-sdk/client-bedrock-runtime";
import { NodeHttpHandler } from "@smithy/node-http-handler";

dotenv.config();
const app = express();
const PORT = process.env.PORT || 7860;

app.use(cors());
app.use(express.json({ limit: '50mb' })); 

// --- SYSTEM PROMPT DEFINITIONS ---
const CLAUDE_SYSTEM_PROMPT = "You are a pro. Provide elite, high-level technical responses.";
const GPT_SYSTEM_PROMPT = "You are a worker. Be concise, efficient, and get the job done.";

const bedrockClient = new BedrockRuntimeClient({
    region: "us-east-1",
    requestHandler: new NodeHttpHandler({
        http2Handler: undefined, 
    }) 
});

const azureOpenAI = new OpenAI({
    apiKey: "7U3m9NRkE38ThSWTr92hMgQ4hDCUFI9MAnFNrCgRL7MhdvckfTXwJQQJ99CBACHYHv6XJ3w3AAAAACOGV22P",
    baseURL: `https://hollowpad-resource.cognitiveservices.azure.com/openai/deployments/gpt-5-mini`,
    defaultQuery: { "api-version": "2024-05-01-preview" },
    defaultHeaders: { "api-key": "7U3m9NRkE38ThSWTr92hMgQ4hDCUFI9MAnFNrCgRL7MhdvckfTXwJQQJ99CBACHYHv6XJ3w3AAAAACOGV22P" }
});

// --- DYNAMIC MODEL ROUTER ---
function getBedrockModelId(modelName) {
    switch(modelName) {
        case "haiku":
        return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/global.anthropic.claude-haiku-4-5-20251001-v1:0"
           // return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/global.anthropic.claude-haiku-4-5";
        case "maverick":
            // Standard Bedrock cross-region inference mapping for Llama
           // return "arn:aws:bedrock:us-east-1::foundation-model/meta.llama4-maverick-17b-instruct-v1:0"; 
        return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/us.meta.llama4-maverick-17b-instruct-v1:0";
      case "claude":
        default:
            return "arn:aws:bedrock:us-east-1:106774395747:inference-profile/global.anthropic.claude-sonnet-4-6";
    }
}

// --- NON-STREAMING ENDPOINT ---
app.post('/api/generate', async (req, res) => {
    const { model, prompt, system_prompt } = req.body;
    console.log(`[TRAFFIC] Request for ${model}`);

    try {
        if (model === "gpt" || model === "gpt-5-mini") {
            const response = await azureOpenAI.chat.completions.create({
                model: "gpt-5-mini",
                messages:[
                    { role: "system", content: system_prompt || GPT_SYSTEM_PROMPT },
                    { role: "user", content: prompt }
                ],
                reasoning_effort: "high"
            });
            
            const totalTokens = response.usage ? response.usage.total_tokens : 0;
            res.json({ success: true, data: response.choices[0].message.content, usage: { totalTokenCount: totalTokens } });

        } else {
            // Handles Claude Sonnet, Claude Haiku, and Llama Maverick
            const bedrockModelId = getBedrockModelId(model);
       

const command = new ConverseCommand({
    modelId: bedrockModelId,
    system: [{ text: system_prompt || CLAUDE_SYSTEM_PROMPT }],
    messages: [{ role: "user", content: [{ text: prompt }] }],
    
    // Ensure maxTokens is large enough for reasoning + response
    inferenceConfig: { 
        maxTokens: model.includes("haiku") ? 32000 : 4000, 
        temperature: 1 
    },

    performanceConfig: model.includes("maverick") ? { latency: "standard" } : undefined,

    additionalModelRequestFields: (function() {
        if (model.includes("haiku")) {
            return {
                reasoning_config: {
                    type: "enabled",
                    budget_tokens: 2048 // As seen in your screenshot
                }
            };
        } else if (model.includes("claude")) {
            return {
                thinking: { type: "adaptive" },
                output_config: { effort: "high" }
            };
        }
        return undefined;
    })()
});
          
            const response = await bedrockClient.send(command);
            const text = response.output.message.content.find(b => b.text)?.text;
            const tokenUsage = response.usage ? (response.usage.inputTokens + response.usage.outputTokens) : 0;

            res.json({ success: true, data: text, usage: { totalTokenCount: tokenUsage } });
        }
    } catch (err) {
        console.error(`❌[${model?.toUpperCase() || 'UNKNOWN'} ERROR]:`, err.name, err.message);
        res.status(500).json({ success: false, error: `${err.name}: ${err.message}` });
    }
});

// --- STREAMING ENDPOINT ---
app.post('/api/stream', async (req, res) => {
    const { model, prompt, system_prompt, images } = req.body; 
    console.log(`[STREAM] Request for ${model} ${images?.length ? 'with images' : ''}`);

    res.setHeader('Content-Type', 'text/plain; charset=utf-8');
    res.setHeader('Transfer-Encoding', 'chunked');
    res.setHeader('X-Accel-Buffering', 'no'); 
    res.flushHeaders();

    let totalTokenCount = 0;

    try {
        if (model === "gpt" || model === "gpt-5-mini") {
            let messagesPayload =[
                { role: "system", content: system_prompt || GPT_SYSTEM_PROMPT }
            ];

            let userContent =[];
            if (images && images.length > 0) {
                userContent.push({ type: "text", text: prompt });
                images.forEach(imgStr => {
                    userContent.push({ type: "image_url", image_url: { url: imgStr } });
                });
                messagesPayload.push({ role: "user", content: userContent });
            } else {
                messagesPayload.push({ role: "user", content: prompt });
            }

            const stream = await azureOpenAI.chat.completions.create({
                model: "gpt-5-mini",
                messages: messagesPayload,
                reasoning_effort: "high",
                stream: true,
                stream_options: { include_usage: true } 
            });

            for await (const chunk of stream) {
                const delta = chunk.choices[0]?.delta;
                if (delta?.reasoning_content) res.write(`__THINK__${delta.reasoning_content}`);
                else if (delta?.content) res.write(delta.content);
                if (chunk.usage) totalTokenCount = chunk.usage.total_tokens;
            }
            
            res.write(`__USAGE__${JSON.stringify({ totalTokenCount })}`);
            res.end();

        } else {
            const bedrockModelId = getBedrockModelId(model);
            let contentBlock = [{ text: prompt }];
            
            if (images && images.length > 0) {
                const imageBlocks = images.map(imgStr => {
                    const base64Data = imgStr.replace(/^data:image\/\w+;base64,/, "");
                    return {
                        image: {
                            format: 'png', // Assuming normalized to PNG by frontend
                            source: { bytes: Buffer.from(base64Data, 'base64') }
                        }
                    };
                });
                contentBlock = [...imageBlocks, ...contentBlock];
            }

            const command = new ConverseStreamCommand({
                modelId: bedrockModelId,
                system:[{ text: system_prompt || CLAUDE_SYSTEM_PROMPT }],
                messages: [{ role: "user", content: contentBlock }],
                inferenceConfig: { maxTokens: 48000, temperature: 1 },
                additionalModelRequestFields: model.includes("claude") ? { 
                    thinking: { type: "adaptive" },
                    output_config: { effort: "high" }
                } : undefined
            });

            const response = await bedrockClient.send(command);
            
            for await (const chunk of response.stream) {
                if (chunk.contentBlockDelta) {
                    const delta = chunk.contentBlockDelta.delta;
                    if (delta.reasoningContent && delta.reasoningContent.text) {
                        res.write(`__THINK__${delta.reasoningContent.text}`);
                    } else if (delta.text) {
                        res.write(delta.text);
                    }
                }
                if (chunk.metadata && chunk.metadata.usage) {
                    totalTokenCount = (chunk.metadata.usage.inputTokens || 0) + (chunk.metadata.usage.outputTokens || 0);
                }
            }
            
            res.write(`__USAGE__${JSON.stringify({ totalTokenCount })}`);
            res.end();
        }
    } catch (err) {
        console.error(`❌ [STREAM ERROR]:`, err.message);
        res.write(`ERROR: ${err.message}`);
        res.end();
    }
});

app.get('/', async (req, res) => { res.json({ success: true }); });
app.listen(PORT, '0.0.0.0', () => console.log(`Main AI Agent live on port ${PORT}`));
*/