Add selectable scopes and models to RAG playground

RAG/public/playground/app.js

@@ -14,6 +14,7 @@ const queryMode = document.getElementById("queryMode");
 const queryIntent = document.getElementById("queryIntent");
 const queryOperation = document.getElementById("queryOperation");
 const answerModel = document.getElementById("answerModel");
+const useModelInRetrieve = document.getElementById("useModelInRetrieve");
 const scopePresetSelect = document.getElementById("scopePresetSelect");
 const scopeSourceRef = document.getElementById("scopeSourceRef");
 const scopeTags = document.getElementById("scopeTags");
@@ -83,6 +84,30 @@ async function loadScopes() {
   }
 }
 
+async function loadAnswerModels() {
+  try {
+    const payload = await fetch("/models/answer").then((response) => response.json());
+    answerModel.innerHTML = "";
+
+    for (const model of payload.models || []) {
+      const option = document.createElement("option");
+      option.value = model;
+      option.textContent = model;
+      answerModel.appendChild(option);
+    }
+
+    if (payload.defaultModel) {
+      answerModel.value = payload.defaultModel;
+    }
+  } catch {
+    answerModel.innerHTML = "";
+    const option = document.createElement("option");
+    option.value = "openai/gpt-4.1-mini";
+    option.textContent = "openai/gpt-4.1-mini";
+    answerModel.appendChild(option);
+  }
+}
+
 scopePresetSelect.addEventListener("change", () => {
   if (!scopePresetSelect.value) {
     return;
@@ -134,6 +159,7 @@ queryButton.addEventListener("click", async () => {
     intent: queryIntent.value,
     query: queryInput.value,
     model: answerModel.value,
+    useModelInRetrieve: useModelInRetrieve.checked,
     scope: {
       sourceRef: scopeSourceRef.value,
       tags: splitTags(scopeTags.value)
@@ -186,3 +212,4 @@ presetCode.addEventListener("click", () => {
 });
 
 loadScopes();
+loadAnswerModels();

RAG/public/playground/index.html

@@ -68,7 +68,9 @@
               </select>
             </label>
             <label>Modelo answer
-              <input id="answerModel" value="openai/gpt-4.1-mini" />
+              <select id="answerModel">
+                <option value="openai/gpt-4.1-mini">openai/gpt-4.1-mini</option>
+              </select>
             </label>
           </div>
 
@@ -91,6 +93,11 @@
             Comparar tambien con respuesta sin RAG
           </label>
 
+          <label class="checkbox">
+            <input type="checkbox" id="useModelInRetrieve" />
+            Usar tambien un modelo para sintetizar el resultado de retrieve
+          </label>
+
           <div class="actions">
             <button id="queryButton">Ejecutar consulta</button>
             <button id="presetDocs" class="secondary">Preset docs</button>

RAG/src/app.ts

@@ -58,6 +58,18 @@ export function createApp() {
     }
   });
 
+  app.get("/models/answer", async (_req, res) => {
+    try {
+      const models = await answerService.listAvailableAnswerModels();
+      res.json({
+        defaultModel: env.answerModel,
+        models
+      });
+    } catch (error) {
+      res.status(500).json({ ok: false, error: error instanceof Error ? error.message : "Unknown answer models error" });
+    }
+  });
+
   app.post("/ingest", async (req, res) => {
     try {
       const result = await ingestService.ingest(req.body);
@@ -72,6 +84,8 @@ export function createApp() {
       const mode = (req.body.mode ?? "auto") as ChunkMode;
       const intent = (req.body.intent ?? "specific") as RetrieveIntent;
       const query = String(req.body.query ?? "");
+      const model = req.body.model ? String(req.body.model) : undefined;
+      const useModel = Boolean(req.body.useModelInRetrieve);
       const scope: RetrieveScope | undefined = req.body.scope
         ? {
             sourceId: req.body.scope.sourceId,
@@ -80,6 +94,15 @@ export function createApp() {
           }
         : undefined;
       const result = await retrieveService.retrieve(mode, intent, query, scope);
+      if (useModel) {
+        const modelSummary = await answerService.summarizeRetrieve(query, result, model);
+        res.json({
+          ...result,
+          model: modelSummary.model,
+          modelSummary: modelSummary.summary
+        });
+        return;
+      }
       res.json(result);
     } catch (error) {
       res.status(500).json({ ok: false, error: error instanceof Error ? error.message : "Unknown retrieve error" });

RAG/src/modules/answer/service.ts

@@ -1,8 +1,16 @@
 import OpenAI from "openai";
 import { env } from "../../config/env.js";
-import type { AnswerResponse, ChunkMode, DirectAnswerResponse, RetrieveIntent, RetrieveScope, RetrievedItem } from "../../shared/types/rag.js";
+import type { AnswerResponse, ChunkMode, DirectAnswerResponse, RetrieveIntent, RetrieveResponse, RetrieveScope, RetrievedItem } from "../../shared/types/rag.js";
 import { RetrieveService } from "../retrieve/service.js";
 
+const answerModelFallbacks = [
+  "openai/gpt-4.1-mini",
+  "qwen/qwen-2.5-72b-instruct",
+  "google/gemini-2.0-flash-001",
+  "anthropic/claude-3.5-haiku",
+  "mistralai/mistral-small-3.1-24b-instruct"
+];
+
 function buildPrompt(query: string, summary: string, items: RetrievedItem[]): string {
   const context = items
     .map((item, index) => [
@@ -104,4 +112,70 @@ export class AnswerService {
       answer: completion.choices[0]?.message?.content?.trim() || "No se pudo generar respuesta."
     };
   }
+
+  async summarizeRetrieve(query: string, retrieved: RetrieveResponse, modelOverride?: string): Promise<{ model: string; summary: string }> {
+    if (!env.answerApiKey) {
+      throw new Error("Missing ANSWER_API_KEY for answer provider");
+    }
+
+    const answerModel = modelOverride?.trim() || env.answerModel;
+    const completion = await this.client.chat.completions.create({
+      model: answerModel,
+      temperature: 0.2,
+      messages: [
+        {
+          role: "system",
+          content: "Eres un sintetizador de contexto RAG. No des una respuesta final al usuario; devuelve una panoramica util y breve del contexto recuperado para que otro agente pueda continuar trabajando con el."
+        },
+        {
+          role: "user",
+          content: [
+            `Consulta original: ${query}`,
+            `Intent: ${retrieved.intent}`,
+            `Resumen base: ${retrieved.summary}`,
+            `Temas: ${retrieved.topics.join(", ")}`,
+            `Puntos criticos: ${retrieved.criticalPoints.join(", ")}`,
+            "Fragmentos recuperados:",
+            retrieved.items.map((item, index) => [
+              `Fuente ${index + 1}: ${item.title}`,
+              item.sectionTitle ? `section_title: ${item.sectionTitle}` : undefined,
+              item.startLine ? `line_range: ${item.startLine}-${item.endLine ?? item.startLine}` : undefined,
+              item.content
+            ].filter(Boolean).join("\n")).join("\n\n---\n\n")
+          ].join("\n\n")
+        }
+      ]
+    });
+
+    return {
+      model: answerModel,
+      summary: completion.choices[0]?.message?.content?.trim() || retrieved.summary
+    };
+  }
+
+  async listAvailableAnswerModels(): Promise<string[]> {
+    try {
+      const response = await fetch(`${env.answerBaseUrl}/models`, {
+        headers: {
+          Authorization: `Bearer ${env.answerApiKey}`
+        }
+      });
+
+      if (!response.ok) {
+        throw new Error(`Model list request failed with ${response.status}`);
+      }
+
+      const payload = await response.json() as { data?: Array<{ id?: string }> };
+      const ids = (payload.data ?? [])
+        .map((item) => item.id)
+        .filter((id): id is string => Boolean(id))
+        .filter((id) => !/embed/i.test(id));
+
+      const preferred = answerModelFallbacks.filter((id) => ids.includes(id));
+      const others = ids.filter((id) => !preferred.includes(id)).sort();
+      return [...preferred, ...others].slice(0, 80);
+    } catch {
+      return answerModelFallbacks;
+    }
+  }
 }

RAG/src/shared/types/rag.ts

@@ -58,7 +58,9 @@ export interface AvailableScope {
 export interface RetrieveResponse {
   mode: ChunkMode;
   intent: RetrieveIntent;
+  model?: string;
   summary: string;
+  modelSummary?: string;
   topics: string[];
   criticalPoints: string[];
   items: RetrievedItem[];