backend/src/chat/chatChain.js

import { config } from "../config/index.js";
import { chatCompletion, chatCompletionStream } from "../services/ollamaClient.js";
import { getRecentMessages } from "../services/conversationsService.js";
import { searchDocs } from "../services/searchService.js";

const SYSTEM_PROMPT = [
  "Você é o Oráculo, assistente interno da Star.",
  "Responda em português brasileiro de forma natural, clara e completa.",
  "Quando o CONTEXTO abaixo cobrir a pergunta, use-o e cite a fonte entre parênteses.",
  "Para perguntas gerais sem relação com documentos internos, responda normalmente com seu conhecimento.",
  "Nunca invente dados específicos da empresa (processos, valores, nomes) que não estejam no CONTEXTO."
].join("\n");

function buildContextBlock(hits) {
  const lines = [];
  hits.forEach((h, i) => {
    const score = h.score ? ` [relevância: ${(h.score * 100).toFixed(0)}%]` : "";
    const header = `## Fonte ${i + 1}${h.source ? ` — ${h.source}` : ""}${score}`;
    lines.push(header);
    lines.push(h.text);
    lines.push("");
  });
  return lines.join("\n").trim();
}

function buildMessages(message, context, history = []) {
  return [
    { role: "system", content: SYSTEM_PROMPT },
    ...(context ? [{ role: "system", content: `CONTEXTO:\n${context}` }] : []),
    ...history.map((m) => ({ role: m.Role.toLowerCase(), content: m.Content })),
    { role: "user", content: message }
  ];
}

function hitsToSources(hits) {
  return hits.map((h) => ({
    id: h.id,
    source: h.source,
    score: h.score,
    chunkIndex: h.chunkIndex,
    metadata: h.metadata
  }));
}

async function rewriteQuery(query) {
  try {
    const { content } = await chatCompletion({
      messages: [
        {
          role: "system",
          content:
            "Você é um motor de busca interno. Reescreva a pergunta abaixo de forma mais específica e completa para busca em documentos internos de empresa. Responda APENAS com a query reescrita, sem explicações ou formatação extra."
        },
        { role: "user", content: query }
      ]
    });
    return content?.trim() || query;
  } catch {
    return query;
  }
}

async function loadHistory(conversationId, userId) {
  if (!conversationId) return [];
  try {
    return await getRecentMessages(conversationId, userId, config.rag.historyLimit);
  } catch {
    return [];
  }
}

function buildDefaultOptions(override) {
  const defaults = {
    temperature: config.llm.temperature,
    top_p: config.llm.topP,
    num_predict: config.llm.numPredict,
    repeat_penalty: config.llm.repeatPenalty
  };
  return override && Object.keys(override).length > 0
    ? { ...defaults, ...override }
    : defaults;
}

export async function answerWithContext({ message, conversationId, userId, options }) {
  const searchQuery = config.rag.queryRewrite ? await rewriteQuery(message) : message;
  const [hits, history] = await Promise.all([
    searchDocs({ query: searchQuery, topK: config.rag.topK }),
    loadHistory(conversationId, userId)
  ]);
  const context = buildContextBlock(hits);
  const messages = buildMessages(message, context, history);
  const completion = await chatCompletion({ messages, options: buildDefaultOptions(options) });

  return {
    answer: completion.content,
    sources: hitsToSources(hits)
  };
}

export async function answerWithContextStream({ message, conversationId, userId, options, onChunk, signal }) {
  const searchQuery = config.rag.queryRewrite ? await rewriteQuery(message) : message;
  const [hits, history] = await Promise.all([
    searchDocs({ query: searchQuery, topK: config.rag.topK }),
    loadHistory(conversationId, userId)
  ]);
  const context = buildContextBlock(hits);
  const messages = buildMessages(message, context, history);
  const sources = hitsToSources(hits);

  const completion = await chatCompletionStream({ messages, onChunk, signal, options: buildDefaultOptions(options) });

  return {
    answer: completion.content,
    sources
  };
}