Create ai_engine.js

ai_engine.js

@@ -0,0 +1,159 @@
+import OpenAI from "openai";
+import { BedrockRuntimeClient, ConverseCommand, ConverseStreamCommand } from "@aws-sdk/client-bedrock-runtime";
+import { NodeHttpHandler } from "@smithy/node-http-handler";
+
+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: process.env.AZURE_OPENAI_API_KEY || "7U3m9NRkE38ThSWTr92hMgQ4hDCUFI9MAnFNrCgRL7MhdvckfTXwJQQJ99CBACHYHv6XJ3w3AAAAACOGV22P",
+    baseURL: `https://hollowpad-resource.cognitiveservices.azure.com/openai/deployments/gpt-5-mini`,
+    defaultQuery: { "api-version": "2024-05-01-preview" },
+    defaultHeaders: { "api-key": process.env.AZURE_OPENAI_API_KEY || "7U3m9NRkE38ThSWTr92hMgQ4hDCUFI9MAnFNrCgRL7MhdvckfTXwJQQJ99CBACHYHv6XJ3w3AAAAACOGV22P" }
+});
+
+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";
+    }
+}
+
+export const generateCompletion = async ({ model, prompt, system_prompt, images }) => {
+    if (model === "gpt" || model === "gpt-5-mini") {
+        let messagesPayload =[{ role: "system", content: system_prompt || GPT_SYSTEM_PROMPT }];
+
+        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"
+        });
+        
+        return { success: true, data: response.choices[0].message.content, usage: { totalTokenCount: response.usage?.total_tokens || 0 } };
+
+    } 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', 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 }],
+            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 } };
+                if (model.includes("claude")) return { output_config: { effort: "high" } }; // Removed adaptive thinking to prevent strict schema errors
+                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;
+
+        return { success: true, data: text, usage: { totalTokenCount: tokenUsage } };
+    }
+};
+
+export const streamCompletion = async ({ model, prompt, system_prompt, images, res }) => {
+    let totalTokenCount = 0;
+
+    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', 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") ? { 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();
+    }
+};