@retrivora-ai/rag-engine 1.0.2 → 1.0.3

package/dist/{chunk-OJNAKSQ2.mjs → chunk-3JR3SAWX.mjs}

@@ -1391,26 +1391,39 @@ var EntityExtractor = class {
   async extract(text) {
     const prompt = `
 Extract entities and relationships from the following text.
-Format the output as JSON with two keys: "nodes" (array of {id, label, properties}) and "edges" (array of {source, target, type, properties}).
-Use the same ID for the same entity.
+Format the output as a JSON object with exactly two keys: "nodes" and "edges".
 
-IMPORTANT: Ensure all property values are valid JSON types (strings, numbers, booleans, or null). 
-DO NOT include mathematical expressions like "4.5/5" as numbers; use strings instead.
+Nodes: {id, label, properties}
+Edges: {source, target, type, properties}
 
-Text:
-"${text}"
+IMPORTANT:
+- Ensure all property values are simple JSON types (string, number, boolean).
+- Do not use mathematical expressions or complex nested objects in properties.
+- If no entities are found, return empty arrays.
+- RESPOND ONLY WITH THE JSON OBJECT.
 
-Output JSON:
+Text to extract from:
+"${text}"
 `;
     const response = await this.llm.chat([
-      { role: "system", content: "You are an expert at knowledge graph extraction. Respond only with valid JSON." },
+      { role: "system", content: "You are a precise knowledge graph extraction engine. You always output valid JSON and nothing else." },
       { role: "user", content: prompt }
-    ], "");
+    ], "", { maxTokens: 2048, temperature: 0.1 });
     try {
-      const cleanJson = response.replace(/```json\n?|\n?```/g, "").trim();
+      const jsonMatch = response.match(/\{[\s\S]*\}/);
+      const cleanJson = jsonMatch ? jsonMatch[0] : response.trim();
       return JSON.parse(cleanJson);
     } catch (e) {
-      console.warn("[EntityExtractor] Failed to parse LLM response as JSON:", response);
+      console.warn(`[EntityExtractor] Failed to parse LLM response. Length: ${response.length} chars.`);
+      if (response.trim().endsWith("}") === false && response.includes('"nodes"')) {
+        try {
+          const partialResponse = response.trim() + (response.includes("[") ? "]}" : "}");
+          const jsonMatch = partialResponse.match(/\{[\s\S]*\}/);
+          if (jsonMatch) return JSON.parse(jsonMatch[0]);
+        } catch (e2) {
+        }
+      }
+      console.warn("[EntityExtractor] Full failing response:", response.substring(0, 200) + "...");
       return { nodes: [], edges: [] };
     }
   }

package/dist/handlers/index.js

@@ -2854,26 +2854,39 @@ var EntityExtractor = class {
   async extract(text) {
     const prompt = `
 Extract entities and relationships from the following text.
-Format the output as JSON with two keys: "nodes" (array of {id, label, properties}) and "edges" (array of {source, target, type, properties}).
-Use the same ID for the same entity.
+Format the output as a JSON object with exactly two keys: "nodes" and "edges".
 
-IMPORTANT: Ensure all property values are valid JSON types (strings, numbers, booleans, or null). 
-DO NOT include mathematical expressions like "4.5/5" as numbers; use strings instead.
+Nodes: {id, label, properties}
+Edges: {source, target, type, properties}
 
-Text:
-"${text}"
+IMPORTANT:
+- Ensure all property values are simple JSON types (string, number, boolean).
+- Do not use mathematical expressions or complex nested objects in properties.
+- If no entities are found, return empty arrays.
+- RESPOND ONLY WITH THE JSON OBJECT.
 
-Output JSON:
+Text to extract from:
+"${text}"
 `;
     const response = await this.llm.chat([
-      { role: "system", content: "You are an expert at knowledge graph extraction. Respond only with valid JSON." },
+      { role: "system", content: "You are a precise knowledge graph extraction engine. You always output valid JSON and nothing else." },
       { role: "user", content: prompt }
-    ], "");
+    ], "", { maxTokens: 2048, temperature: 0.1 });
     try {
-      const cleanJson = response.replace(/```json\n?|\n?```/g, "").trim();
+      const jsonMatch = response.match(/\{[\s\S]*\}/);
+      const cleanJson = jsonMatch ? jsonMatch[0] : response.trim();
       return JSON.parse(cleanJson);
     } catch (e) {
-      console.warn("[EntityExtractor] Failed to parse LLM response as JSON:", response);
+      console.warn(`[EntityExtractor] Failed to parse LLM response. Length: ${response.length} chars.`);
+      if (response.trim().endsWith("}") === false && response.includes('"nodes"')) {
+        try {
+          const partialResponse = response.trim() + (response.includes("[") ? "]}" : "}");
+          const jsonMatch = partialResponse.match(/\{[\s\S]*\}/);
+          if (jsonMatch) return JSON.parse(jsonMatch[0]);
+        } catch (e2) {
+        }
+      }
+      console.warn("[EntityExtractor] Full failing response:", response.substring(0, 200) + "...");
       return { nodes: [], edges: [] };
     }
   }

package/dist/handlers/index.mjs

@@ -4,7 +4,7 @@ import {
   createIngestHandler,
   createStreamHandler,
   createUploadHandler
-} from "../chunk-OJNAKSQ2.mjs";
+} from "../chunk-3JR3SAWX.mjs";
 import "../chunk-YLTMFW4M.mjs";
 import "../chunk-X4TOT24V.mjs";
 export {

package/dist/server.js

@@ -2936,26 +2936,39 @@ var EntityExtractor = class {
   async extract(text) {
     const prompt = `
 Extract entities and relationships from the following text.
-Format the output as JSON with two keys: "nodes" (array of {id, label, properties}) and "edges" (array of {source, target, type, properties}).
-Use the same ID for the same entity.
+Format the output as a JSON object with exactly two keys: "nodes" and "edges".
 
-IMPORTANT: Ensure all property values are valid JSON types (strings, numbers, booleans, or null). 
-DO NOT include mathematical expressions like "4.5/5" as numbers; use strings instead.
+Nodes: {id, label, properties}
+Edges: {source, target, type, properties}
 
-Text:
-"${text}"
+IMPORTANT:
+- Ensure all property values are simple JSON types (string, number, boolean).
+- Do not use mathematical expressions or complex nested objects in properties.
+- If no entities are found, return empty arrays.
+- RESPOND ONLY WITH THE JSON OBJECT.
 
-Output JSON:
+Text to extract from:
+"${text}"
 `;
     const response = await this.llm.chat([
-      { role: "system", content: "You are an expert at knowledge graph extraction. Respond only with valid JSON." },
+      { role: "system", content: "You are a precise knowledge graph extraction engine. You always output valid JSON and nothing else." },
       { role: "user", content: prompt }
-    ], "");
+    ], "", { maxTokens: 2048, temperature: 0.1 });
     try {
-      const cleanJson = response.replace(/```json\n?|\n?```/g, "").trim();
+      const jsonMatch = response.match(/\{[\s\S]*\}/);
+      const cleanJson = jsonMatch ? jsonMatch[0] : response.trim();
       return JSON.parse(cleanJson);
     } catch (e) {
-      console.warn("[EntityExtractor] Failed to parse LLM response as JSON:", response);
+      console.warn(`[EntityExtractor] Failed to parse LLM response. Length: ${response.length} chars.`);
+      if (response.trim().endsWith("}") === false && response.includes('"nodes"')) {
+        try {
+          const partialResponse = response.trim() + (response.includes("[") ? "]}" : "}");
+          const jsonMatch = partialResponse.match(/\{[\s\S]*\}/);
+          if (jsonMatch) return JSON.parse(jsonMatch[0]);
+        } catch (e2) {
+        }
+      }
+      console.warn("[EntityExtractor] Full failing response:", response.substring(0, 200) + "...");
       return { nodes: [], edges: [] };
     }
   }

package/dist/server.mjs

@@ -34,7 +34,7 @@ import {
   createIngestHandler,
   createUploadHandler,
   getRagConfig
-} from "./chunk-OJNAKSQ2.mjs";
+} from "./chunk-3JR3SAWX.mjs";
 import "./chunk-YLTMFW4M.mjs";
 import {
   PineconeProvider

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@retrivora-ai/rag-engine",
-  "version": "1.0.2",
+  "version": "1.0.3",
   "description": "Retrivora AI is a plug-and-play AI engine for RAG chat experiences — generic vector DB + LLM provider, embeddable or standalone.",
   "author": "Abhinav Alkuchi",
   "license": "MIT",

package/src/rag/EntityExtractor.ts

@@ -17,29 +17,46 @@ export class EntityExtractor {
   async extract(text: string): Promise<{ nodes: GraphNode[]; edges: Edge[] }> {
     const prompt = `
 Extract entities and relationships from the following text.
-Format the output as JSON with two keys: "nodes" (array of {id, label, properties}) and "edges" (array of {source, target, type, properties}).
-Use the same ID for the same entity.
+Format the output as a JSON object with exactly two keys: "nodes" and "edges".
 
-IMPORTANT: Ensure all property values are valid JSON types (strings, numbers, booleans, or null). 
-DO NOT include mathematical expressions like "4.5/5" as numbers; use strings instead.
+Nodes: {id, label, properties}
+Edges: {source, target, type, properties}
 
-Text:
-"${text}"
+IMPORTANT:
+- Ensure all property values are simple JSON types (string, number, boolean).
+- Do not use mathematical expressions or complex nested objects in properties.
+- If no entities are found, return empty arrays.
+- RESPOND ONLY WITH THE JSON OBJECT.
 
-Output JSON:
+Text to extract from:
+"${text}"
 `;
 
     const response = await this.llm.chat([
-      { role: 'system', content: 'You are an expert at knowledge graph extraction. Respond only with valid JSON.' },
+      { role: 'system', content: 'You are a precise knowledge graph extraction engine. You always output valid JSON and nothing else.' },
       { role: 'user', content: prompt }
-    ], '');
+    ], '', { maxTokens: 2048, temperature: 0.1 });
 
     try {
-      // Clean up LLM response (sometimes it adds markdown blocks)
-      const cleanJson = response.replace(/```json\n?|\n?```/g, '').trim();
+      // 1. Try to find the JSON block using a regex
+      const jsonMatch = response.match(/\{[\s\S]*\}/);
+      const cleanJson = jsonMatch ? jsonMatch[0] : response.trim();
+      
       return JSON.parse(cleanJson);
     } catch {
-      console.warn('[EntityExtractor] Failed to parse LLM response as JSON:', response);
+      console.warn(`[EntityExtractor] Failed to parse LLM response. Length: ${response.length} chars.`);
+      
+      // Attempt very basic "fix" if it looks like it was truncated
+      if (response.trim().endsWith('}') === false && response.includes('"nodes"')) {
+        try {
+          // This is a last-resort attempt to see if we can get anything out of a truncated response
+          const partialResponse = response.trim() + (response.includes('[') ? ']}' : '}');
+          const jsonMatch = partialResponse.match(/\{[\s\S]*\}/);
+          if (jsonMatch) return JSON.parse(jsonMatch[0]);
+        } catch { /* ignore */ }
+      }
+
+      console.warn('[EntityExtractor] Full failing response:', response.substring(0, 200) + '...');
       return { nodes: [], edges: [] };
     }
   }