@mastra/rag 1.0.8-alpha.0 → 1.0.9-alpha.0

package/dist/index.js

@@ -1,9 +1,9 @@
+import { Agent } from '@mastra/core/agent';
 import { randomUUID, createHash } from 'crypto';
 import { z } from 'zod';
 import { parse } from 'node-html-better-parser';
 import { encodingForModel, getEncoding } from 'js-tiktoken';
 import { Big } from 'big.js';
-import { Agent } from '@mastra/core/agent';
 import { createSimilarityPrompt } from '@mastra/core/relevance';
 import ZeroEntropy from 'zeroentropy';
 import { createTool } from '@mastra/core/tools';
@@ -2285,7 +2285,7 @@ var openaiChatChunkSchema = z.union([
   openaiErrorDataSchema
 ]);
 function isReasoningModel(modelId) {
-  return modelId.startsWith("o");
+  return modelId.startsWith("o") || modelId.startsWith("gpt-5");
 }
 function isAudioModel(modelId) {
   return modelId.startsWith("gpt-4o-audio-preview");
@@ -3808,7 +3808,7 @@ function isErrorChunk(chunk) {
   return chunk.type === "error";
 }
 function getResponsesModelConfig(modelId) {
-  if (modelId.startsWith("o")) {
+  if (modelId.startsWith("o") || modelId.startsWith("gpt-5")) {
     if (modelId.startsWith("o1-mini") || modelId.startsWith("o1-preview")) {
       return {
         isReasoningModel: true,
@@ -4134,64 +4134,65 @@ var TitleExtractor = class extends BaseExtractor {
     for (const [key, nodes] of Object.entries(nodesByDocument)) {
       const titleCandidates = await this.getTitlesCandidates(nodes);
       const combinedTitles = titleCandidates.join(", ");
-      const completion = await this.llm.doGenerate({
-        inputFormat: "messages",
-        mode: { type: "regular" },
-        prompt: [
-          {
-            role: "user",
-            content: [
-              {
-                type: "text",
-                text: this.combineTemplate.format({
-                  context: combinedTitles
-                })
-              }
-            ]
-          }
-        ]
-      });
       let title = "";
-      if (typeof completion.text === "string") {
-        title = completion.text.trim();
+      if (this.llm.specificationVersion === "v2") {
+        const miniAgent = new Agent({
+          model: this.llm,
+          name: "title-extractor",
+          instructions: "You are a title extractor. You are given a list of nodes and you need to extract the title from the nodes."
+        });
+        const result = await miniAgent.generateVNext(
+          [{ role: "user", content: this.combineTemplate.format({ context: combinedTitles }) }],
+          { format: "mastra" }
+        );
+        title = result.text;
       } else {
-        console.warn("Title extraction LLM output was not a string:", completion.text);
+        const miniAgent = new Agent({
+          model: this.llm,
+          name: "title-extractor",
+          instructions: "You are a title extractor. You are given a list of nodes and you need to extract the title from the nodes."
+        });
+        const result = await miniAgent.generate([
+          { role: "user", content: this.combineTemplate.format({ context: combinedTitles }) }
+        ]);
+        title = result.text;
+      }
+      if (!title) {
+        console.warn("Title extraction LLM output returned empty");
       }
       titlesByDocument[key] = title;
     }
     return titlesByDocument;
   }
   async getTitlesCandidates(nodes) {
+    const miniAgent = new Agent({
+      model: this.llm,
+      name: "titles-candidates-extractor",
+      instructions: "You are a titles candidates extractor. You are given a list of nodes and you need to extract the titles candidates from the nodes."
+    });
     const titleJobs = nodes.map(async (node) => {
-      const completion = await this.llm.doGenerate({
-        inputFormat: "messages",
-        mode: { type: "regular" },
-        prompt: [
-          {
-            role: "user",
-            content: [
-              {
-                type: "text",
-                text: this.nodeTemplate.format({
-                  context: node.getContent()
-                })
-              }
-            ]
-          }
-        ]
-      });
-      if (typeof completion.text === "string") {
-        return completion.text.trim();
+      let completion;
+      if (this.llm.specificationVersion === "v2") {
+        const result = await miniAgent.generateVNext(
+          [{ role: "user", content: this.nodeTemplate.format({ context: node.getContent() }) }],
+          { format: "mastra" }
+        );
+        completion = result.text;
       } else {
-        console.warn("Title candidate extraction LLM output was not a string:", completion.text);
+        const result = await miniAgent.generate([
+          { role: "user", content: this.nodeTemplate.format({ context: node.getContent() }) }
+        ]);
+        completion = result.text;
+      }
+      if (!completion) {
+        console.warn("Title candidate extraction LLM output returned empty");
         return "";
       }
+      return completion.trim();
     });
     return await Promise.all(titleJobs);
   }
 };
-
-// src/document/extractors/summary.ts
 var SummaryExtractor = class extends BaseExtractor {
   llm;
   summaries;
@@ -4231,21 +4232,22 @@ var SummaryExtractor = class extends BaseExtractor {
     const prompt = this.promptTemplate.format({
       context
     });
-    const result = await this.llm.doGenerate({
-      inputFormat: "messages",
-      mode: { type: "regular" },
-      prompt: [
-        {
-          role: "user",
-          content: [{ type: "text", text: prompt }]
-        }
-      ]
+    const miniAgent = new Agent({
+      model: this.llm,
+      name: "summary-extractor",
+      instructions: "You are a summary extractor. You are given a node and you need to extract the summary from the node."
     });
     let summary = "";
-    if (typeof result.text === "string") {
-      summary = result.text.trim();
+    if (this.llm.specificationVersion === "v2") {
+      const result = await miniAgent.generateVNext([{ role: "user", content: prompt }], { format: "mastra" });
+      summary = result.text;
     } else {
-      console.warn("Summary extraction LLM output was not a string:", result.text);
+      const result = await miniAgent.generate([{ role: "user", content: prompt }]);
+      summary = result.text;
+    }
+    if (!summary) {
+      console.warn("Summary extraction LLM output returned empty");
+      return "";
     }
     return summary.replace(STRIP_REGEX, "");
   }
@@ -4272,8 +4274,6 @@ var SummaryExtractor = class extends BaseExtractor {
     return metadataList;
   }
 };
-
-// src/document/extractors/questions.ts
 var QuestionsAnsweredExtractor = class extends BaseExtractor {
   llm;
   questions = 5;
@@ -4317,26 +4317,24 @@ var QuestionsAnsweredExtractor = class extends BaseExtractor {
       context: contextStr,
       numQuestions: this.questions.toString()
     });
-    const questions = await this.llm.doGenerate({
-      inputFormat: "messages",
-      mode: { type: "regular" },
-      prompt: [
-        {
-          role: "user",
-          content: [{ type: "text", text: prompt }]
-        }
-      ]
+    const miniAgent = new Agent({
+      model: this.llm,
+      name: "question-extractor",
+      instructions: "You are a question extractor. You are given a node and you need to extract the questions from the node."
     });
-    let result = "";
-    try {
-      if (typeof questions.text === "string") {
-        result = questions.text.replace(STRIP_REGEX, "").trim();
-      } else {
-        console.warn("Question extraction LLM output was not a string:", questions.text);
-      }
-    } catch (err) {
-      console.warn("Question extraction failed:", err);
+    let questionsText = "";
+    if (this.llm.specificationVersion === "v2") {
+      const result2 = await miniAgent.generateVNext([{ role: "user", content: prompt }], { format: "mastra" });
+      questionsText = result2.text;
+    } else {
+      const result2 = await miniAgent.generate([{ role: "user", content: prompt }]);
+      questionsText = result2.text;
+    }
+    if (!questionsText) {
+      console.warn("Question extraction LLM output returned empty");
+      return { questionsThisExcerptCanAnswer: "" };
     }
+    const result = questionsText.replace(STRIP_REGEX, "").trim();
     return {
       questionsThisExcerptCanAnswer: result
     };
@@ -4351,8 +4349,6 @@ var QuestionsAnsweredExtractor = class extends BaseExtractor {
     return results;
   }
 };
-
-// src/document/extractors/keywords.ts
 var KeywordExtractor = class extends BaseExtractor {
   llm;
   keywords = 5;
@@ -4393,33 +4389,43 @@ var KeywordExtractor = class extends BaseExtractor {
     }
     let keywords = "";
     try {
-      const completion = await this.llm.doGenerate({
-        inputFormat: "messages",
-        mode: { type: "regular" },
-        prompt: [
+      const miniAgent = new Agent({
+        model: this.llm,
+        name: "keyword-extractor",
+        instructions: "You are a keyword extractor. You are given a node and you need to extract the keywords from the node."
+      });
+      if (this.llm.specificationVersion === "v2") {
+        const result = await miniAgent.generateVNext(
+          [
+            {
+              role: "user",
+              content: this.promptTemplate.format({
+                context: node.getContent(),
+                maxKeywords: this.keywords.toString()
+              })
+            }
+          ],
+          { format: "mastra" }
+        );
+        keywords = result.text;
+      } else {
+        const result = await miniAgent.generate([
           {
             role: "user",
-            content: [
-              {
-                type: "text",
-                text: this.promptTemplate.format({
-                  context: node.getContent(),
-                  maxKeywords: this.keywords.toString()
-                })
-              }
-            ]
+            content: this.promptTemplate.format({ context: node.getContent(), maxKeywords: this.keywords.toString() })
           }
-        ]
-      });
-      if (typeof completion.text === "string") {
-        keywords = completion.text.trim();
-      } else {
-        console.warn("Keyword extraction LLM output was not a string:", completion.text);
+        ]);
+        keywords = result.text;
+      }
+      if (!keywords) {
+        console.warn("Keyword extraction LLM output returned empty");
+        return { excerptKeywords: "" };
       }
+      return { excerptKeywords: keywords.trim() };
     } catch (err) {
       console.warn("Keyword extraction failed:", err);
+      return { excerptKeywords: "" };
     }
-    return { excerptKeywords: keywords };
   }
   /**
    *