Research Agent MCP

import { openai } from "@ai-sdk/openai" import { createDataStreamResponse, type JSONValue, streamText } from "ai" import type { NextRequest } from "next/server" import { OPENAI_SEARCH_SUGGESTIONS_PROMPT } from "@/lib/system-prompt" import type { ModelMessage, ModelConfig } from "@/lib/types" import { DEFAULT_MODEL_CONFIG } from "@/lib/models" export const runtime = "nodejs" function extractSearchSuggestionsFromText( text: string, ): { searchTerms: string[]; confidence: number; reasoning: string } | null { try { const regex = /```SEARCH_TERMS_JSON\s*({[\s\S]*?})\s*```/ const match = text.match(regex) if (match && match[1]) { const jsonStr = match[1].trim() const searchSuggestions = JSON.parse(jsonStr) if ( searchSuggestions && Array.isArray(searchSuggestions.searchTerms) && typeof searchSuggestions.confidence === "number" && typeof searchSuggestions.reasoning === "string" ) { return { searchTerms: searchSuggestions.searchTerms, confidence: searchSuggestions.confidence, reasoning: searchSuggestions.reasoning, } } } return null } catch (error) { console.error("Error extracting search suggestions:", error) return null } } function removeSearchTermsJson(text: string): string { const pattern = /```SEARCH_TERMS_JSON\s*({[\s\S]*?})\s*```/g return text.replace(pattern, "").trim() } function validateMessages(messages: any[]): ModelMessage[] { return messages .filter((message) => { const isValid = message && typeof message === "object" && (message.role === "user" || message.role === "assistant" || message.role === "system") && typeof message.content === "string" if (!isValid) { console.warn("Invalid message format detected and filtered out:", message) } return isValid }) .map((message) => ({ role: message.role, content: message.content, ...(message.id && { id: message.id }), ...(message.name && { name: message.name }), })) } export async function POST(req: NextRequest) { try { const body = await req.json() if (!body || !Array.isArray(body.messages)) { return new Response(JSON.stringify({ error: "Invalid request body. Expected messages array." }), { status: 400, headers: { "Content-Type": "application/json" }, }) } let selectedModel = "gpt-4.1-mini" if (body.model && typeof body.model === "string") { selectedModel = body.model } else { const modelCookie = req.cookies.get("selectedOpenAIModel") if (modelCookie) { selectedModel = modelCookie.value } } const modelConfig: ModelConfig = body.modelConfig || DEFAULT_MODEL_CONFIG const reasoningType = body.reasoningType || "Reasoning" const validatedMessages = validateMessages(body.messages) return createDataStreamResponse({ execute: async (dataStream) => { console.log("Starting OpenAI stream execution with model:", selectedModel, "reasoningType:", reasoningType) let fullText = "" const streamOptions = { model: openai.responses(selectedModel), messages: validatedMessages, system: OPENAI_SEARCH_SUGGESTIONS_PROMPT, temperature: modelConfig.temperature, topP: modelConfig.topP, maxTokens: modelConfig.maxTokens, maxSteps: 5, // @ts-expect-error onChunk: ({ chunk }) => { if (chunk.type === "text-delta") { fullText += chunk.textDelta dataStream.writeData({ type: "text-delta", text: chunk.textDelta }) } else { //if (chunk.type === 'reasoning') { process.stdout.write('\x1b[34m' + chunk.textDelta + '\x1b[0m'); } }, // @ts-expect-error onFinish: ({ reasoning, sources, usage, finishReason }) => { if (sources && sources.length > 0) { const formattedSources = sources // @ts-expect-error .filter((source) => source.sourceType === "url") // @ts-expect-error .map((source) => ({ url: source.url, title: source.title || new URL(source.url).hostname, })) if (formattedSources.length > 0) { dataStream.writeData({ type: "sources", sources: formattedSources }) } } const searchSuggestions = extractSearchSuggestionsFromText(fullText) if (searchSuggestions) { dataStream.writeData({ type: "searchSuggestions", searchSuggestions: searchSuggestions.searchTerms, confidence: searchSuggestions.confidence, reasoning: searchSuggestions.reasoning, }) } const cleanedText = removeSearchTermsJson(fullText) dataStream.writeData({ type: "cleaned-text", text: cleanedText, messageId: Date.now().toString(), }) if (usage) { dataStream.writeData({ type: "usage", usage: { prompt_tokens: usage.promptTokens, completion_tokens: usage.completionTokens, total_tokens: usage.totalTokens, finishReason: finishReason || "unknown", }, }) } if (reasoning) { console.log("OpenAI reasoning:", reasoning) } dataStream.writeData({ type: "model-config", config: modelConfig, } as unknown as JSONValue) dataStream.writeData({ type: "selected-model", model: selectedModel, } as unknown as JSONValue) dataStream.writeData({ type: "reasoning-type", reasoning: reasoningType, } as unknown as JSONValue) }, } if (reasoningType === "Intelligence") { Object.assign(streamOptions, { tools: { web_search_preview: openai.tools.webSearchPreview({ searchContextSize: "medium", }), }, toolChoice: { type: "tool", toolName: "web_search_preview" }, }) } if (reasoningType === "Thinking") { Object.assign(streamOptions, { providerOptions: { openai: { reasoningEffort: 'low' }, // This parameter can be set to `'low'`, `'medium'`, or `'high'` to adjust how much time and computation the model spends on internal reasoning before producing a response. reasoningSummary: "auto", }, }) } const result = streamText(streamOptions) result.mergeIntoDataStream(dataStream) }, onError: (error) => { console.error("Error in OpenAI stream:", error) return error instanceof Error ? error.message : String(error) }, }) } catch (error) { console.error("Error in OpenAI chat API:", error) return new Response(JSON.stringify({ error: "Failed to process OpenAI chat request" }), { status: 500, headers: { "Content-Type": "application/json" }, }) } }