@retrivora-ai/rag-engine 1.0.1 → 1.0.3

package/dist/{MongoDBProvider-Z6ALOVDN.mjs → MongoDBProvider-BO2Y5DRR.mjs}

@@ -1,6 +1,6 @@
 import {
   MongoDBProvider
-} from "./chunk-CD6TSNL4.mjs";
+} from "./chunk-73I6VWU3.mjs";
 import "./chunk-IMP6FUCY.mjs";
 import "./chunk-X4TOT24V.mjs";
 export {

package/dist/{chunk-5W2YWFT3.mjs → chunk-3JR3SAWX.mjs}

@@ -1088,7 +1088,7 @@ var ProviderRegistry = class {
         return PostgreSQLProvider;
       }
       case "mongodb": {
-        const { MongoDBProvider } = await import("./MongoDBProvider-Z6ALOVDN.mjs");
+        const { MongoDBProvider } = await import("./MongoDBProvider-BO2Y5DRR.mjs");
         return MongoDBProvider;
       }
       case "milvus": {
@@ -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: [] };
     }
   }
@@ -1977,6 +1990,7 @@ var Pipeline = class {
   async initialize() {
     var _a;
     if (this.initialised) return;
+    console.log(`[Pipeline] Initializing with provider: ${this.config.vectorDb.provider}...`);
     this.vectorDB = await ProviderRegistry.createVectorProvider(this.config.vectorDb);
     const { llmProvider, embeddingProvider } = await EmbeddingStrategyResolver.resolve(
       this.config.llm,
@@ -2361,29 +2375,9 @@ var ProviderHealthCheck = class {
 
 // src/core/VectorPlugin.ts
 var VectorPlugin = class {
-  /**
-   * Initializes the plugin with the host configuration.
-   * @param hostConfig - Configuration object passed from the host application.
-   * @throws Error if configuration is invalid or providers are unhealthy
-   *
-   * @example
-   * const plugin = new VectorPlugin({
-   *   projectId: 'my-app',
-   *   vectorDb: {
-   *     provider: 'pinecone',
-   *     indexName: 'my-index',
-   *     options: { apiKey: process.env.PINECONE_API_KEY }
-   *   },
-   *   llm: {
-   *     provider: 'openai',
-   *     model: 'gpt-4o',
-   *     apiKey: process.env.OPENAI_API_KEY
-   *   }
-   * });
-   */
   constructor(hostConfig) {
     this.config = ConfigResolver.resolve(hostConfig);
-    ConfigValidator.validateAndThrow(this.config);
+    this.validationPromise = ConfigValidator.validateAndThrow(this.config);
     this.pipeline = new Pipeline(this.config);
   }
   /**
@@ -2409,6 +2403,7 @@ var VectorPlugin = class {
    * Run a chat query.
    */
   async chat(message, history = [], namespace) {
+    await this.validationPromise;
     return this.pipeline.ask(message, history, namespace);
   }
   /**
@@ -2416,6 +2411,7 @@ var VectorPlugin = class {
    */
   chatStream(_0) {
     return __asyncGenerator(this, arguments, function* (message, history = [], namespace) {
+      yield new __await(this.validationPromise);
       yield* __yieldStar(this.pipeline.askStream(message, history, namespace));
     });
   }
@@ -2423,6 +2419,7 @@ var VectorPlugin = class {
    * Ingest documents into the vector database.
    */
   async ingest(documents, namespace) {
+    await this.validationPromise;
     return this.pipeline.ingest(documents, namespace);
   }
 };
@@ -2508,33 +2505,43 @@ function createStreamHandler(config) {
   const plugin = new VectorPlugin(config);
   return async function POST(req) {
     try {
-      const { message, history = [], namespace } = await req.json();
+      const body = await req.json();
+      const { message, history = [], namespace } = body;
       const encoder = new TextEncoder();
       const stream = new ReadableStream({
         async start(controller) {
-          const pipelineStream = plugin.chatStream(message, history, namespace);
           try {
-            for (var iter = __forAwait(pipelineStream), more, temp, error; more = !(temp = await iter.next()).done; more = false) {
-              const chunk = temp.value;
-              if (typeof chunk === "string") {
-                controller.enqueue(encoder.encode(chunk));
-              } else {
-                controller.enqueue(encoder.encode(`
+            const pipelineStream = plugin.chatStream(message, history, namespace);
+            try {
+              for (var iter = __forAwait(pipelineStream), more, temp, error; more = !(temp = await iter.next()).done; more = false) {
+                const chunk = temp.value;
+                if (typeof chunk === "string") {
+                  controller.enqueue(encoder.encode(chunk));
+                } else {
+                  controller.enqueue(encoder.encode(`
 
 __METADATA__${JSON.stringify(chunk)}`));
+                }
               }
-            }
-          } catch (temp) {
-            error = [temp];
-          } finally {
-            try {
-              more && (temp = iter.return) && await temp.call(iter);
+            } catch (temp) {
+              error = [temp];
             } finally {
-              if (error)
-                throw error[0];
+              try {
+                more && (temp = iter.return) && await temp.call(iter);
+              } finally {
+                if (error)
+                  throw error[0];
+              }
             }
+            controller.close();
+          } catch (streamError) {
+            console.error("[createStreamHandler] Stream processing error:", streamError);
+            const errorMessage = streamError instanceof Error ? streamError.message : String(streamError);
+            controller.enqueue(encoder.encode(`
+
+__ERROR__${JSON.stringify({ error: errorMessage })}`));
+            controller.close();
           }
-          controller.close();
         }
       });
       return new Response(stream, {

package/dist/{chunk-CD6TSNL4.mjs → chunk-73I6VWU3.mjs}

@@ -14,7 +14,7 @@ var MongoDBProvider = class extends BaseVectorProvider {
     if (!opts.uri || !opts.database || !opts.collection) {
       throw new Error("[MongoDBProvider] options uri, database, and collection are required.");
     }
-    this.client = new MongoClient(opts.uri);
+    this.client = new MongoClient(opts.uri, { serverSelectionTimeoutMS: 5e3 });
     this.dbName = opts.database;
     this.collectionName = opts.collection;
     this.embeddingKey = opts.embeddingKey || "embedding";

package/dist/handlers/index.js

@@ -546,7 +546,7 @@ var init_MongoDBProvider = __esm({
         if (!opts.uri || !opts.database || !opts.collection) {
           throw new Error("[MongoDBProvider] options uri, database, and collection are required.");
         }
-        this.client = new import_mongodb.MongoClient(opts.uri);
+        this.client = new import_mongodb.MongoClient(opts.uri, { serverSelectionTimeoutMS: 5e3 });
         this.dbName = opts.database;
         this.collectionName = opts.collection;
         this.embeddingKey = opts.embeddingKey || "embedding";
@@ -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: [] };
     }
   }
@@ -3434,6 +3447,7 @@ var Pipeline = class {
   async initialize() {
     var _a;
     if (this.initialised) return;
+    console.log(`[Pipeline] Initializing with provider: ${this.config.vectorDb.provider}...`);
     this.vectorDB = await ProviderRegistry.createVectorProvider(this.config.vectorDb);
     const { llmProvider, embeddingProvider } = await EmbeddingStrategyResolver.resolve(
       this.config.llm,
@@ -3818,29 +3832,9 @@ var ProviderHealthCheck = class {
 
 // src/core/VectorPlugin.ts
 var VectorPlugin = class {
-  /**
-   * Initializes the plugin with the host configuration.
-   * @param hostConfig - Configuration object passed from the host application.
-   * @throws Error if configuration is invalid or providers are unhealthy
-   *
-   * @example
-   * const plugin = new VectorPlugin({
-   *   projectId: 'my-app',
-   *   vectorDb: {
-   *     provider: 'pinecone',
-   *     indexName: 'my-index',
-   *     options: { apiKey: process.env.PINECONE_API_KEY }
-   *   },
-   *   llm: {
-   *     provider: 'openai',
-   *     model: 'gpt-4o',
-   *     apiKey: process.env.OPENAI_API_KEY
-   *   }
-   * });
-   */
   constructor(hostConfig) {
     this.config = ConfigResolver.resolve(hostConfig);
-    ConfigValidator.validateAndThrow(this.config);
+    this.validationPromise = ConfigValidator.validateAndThrow(this.config);
     this.pipeline = new Pipeline(this.config);
   }
   /**
@@ -3866,6 +3860,7 @@ var VectorPlugin = class {
    * Run a chat query.
    */
   async chat(message, history = [], namespace) {
+    await this.validationPromise;
     return this.pipeline.ask(message, history, namespace);
   }
   /**
@@ -3873,6 +3868,7 @@ var VectorPlugin = class {
    */
   chatStream(_0) {
     return __asyncGenerator(this, arguments, function* (message, history = [], namespace) {
+      yield new __await(this.validationPromise);
       yield* __yieldStar(this.pipeline.askStream(message, history, namespace));
     });
   }
@@ -3880,6 +3876,7 @@ var VectorPlugin = class {
    * Ingest documents into the vector database.
    */
   async ingest(documents, namespace) {
+    await this.validationPromise;
     return this.pipeline.ingest(documents, namespace);
   }
 };
@@ -3965,33 +3962,43 @@ function createStreamHandler(config) {
   const plugin = new VectorPlugin(config);
   return async function POST(req) {
     try {
-      const { message, history = [], namespace } = await req.json();
+      const body = await req.json();
+      const { message, history = [], namespace } = body;
       const encoder = new TextEncoder();
       const stream = new ReadableStream({
         async start(controller) {
-          const pipelineStream = plugin.chatStream(message, history, namespace);
           try {
-            for (var iter = __forAwait(pipelineStream), more, temp, error; more = !(temp = await iter.next()).done; more = false) {
-              const chunk = temp.value;
-              if (typeof chunk === "string") {
-                controller.enqueue(encoder.encode(chunk));
-              } else {
-                controller.enqueue(encoder.encode(`
+            const pipelineStream = plugin.chatStream(message, history, namespace);
+            try {
+              for (var iter = __forAwait(pipelineStream), more, temp, error; more = !(temp = await iter.next()).done; more = false) {
+                const chunk = temp.value;
+                if (typeof chunk === "string") {
+                  controller.enqueue(encoder.encode(chunk));
+                } else {
+                  controller.enqueue(encoder.encode(`
 
 __METADATA__${JSON.stringify(chunk)}`));
+                }
               }
-            }
-          } catch (temp) {
-            error = [temp];
-          } finally {
-            try {
-              more && (temp = iter.return) && await temp.call(iter);
+            } catch (temp) {
+              error = [temp];
             } finally {
-              if (error)
-                throw error[0];
+              try {
+                more && (temp = iter.return) && await temp.call(iter);
+              } finally {
+                if (error)
+                  throw error[0];
+              }
             }
+            controller.close();
+          } catch (streamError) {
+            console.error("[createStreamHandler] Stream processing error:", streamError);
+            const errorMessage = streamError instanceof Error ? streamError.message : String(streamError);
+            controller.enqueue(encoder.encode(`
+
+__ERROR__${JSON.stringify({ error: errorMessage })}`));
+            controller.close();
           }
-          controller.close();
         }
       });
       return new Response(stream, {

package/dist/handlers/index.mjs

@@ -4,7 +4,7 @@ import {
   createIngestHandler,
   createStreamHandler,
   createUploadHandler
-} from "../chunk-5W2YWFT3.mjs";
+} from "../chunk-3JR3SAWX.mjs";
 import "../chunk-YLTMFW4M.mjs";
 import "../chunk-X4TOT24V.mjs";
 export {

package/dist/index.js

@@ -118,7 +118,7 @@ function MessageBubble({
       className: `relative px-4 py-3 rounded-2xl text-sm leading-relaxed shadow-sm dark:shadow-lg ${isUser ? "text-white rounded-tr-sm" : "bg-white dark:bg-white/5 border border-slate-200 dark:border-white/10 text-slate-800 dark:text-white/90 rounded-tl-sm"}`,
       style: isUser ? { background: `linear-gradient(135deg, ${primaryColor}, ${accentColor})` } : {}
     },
-    isStreaming ? /* @__PURE__ */ import_react2.default.createElement("span", { className: "flex items-center gap-1" }, /* @__PURE__ */ import_react2.default.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:0ms]" }), /* @__PURE__ */ import_react2.default.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:150ms]" }), /* @__PURE__ */ import_react2.default.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:300ms]" })) : /* @__PURE__ */ import_react2.default.createElement("div", { className: `prose prose-sm max-w-none ${isUser ? "prose-invert" : "dark:prose-invert"}` }, /* @__PURE__ */ import_react2.default.createElement(import_react_markdown.default, { remarkPlugins: [import_remark_gfm.default] }, message.content))
+    isStreaming && !message.content ? /* @__PURE__ */ import_react2.default.createElement("span", { className: "flex items-center gap-1 py-1" }, /* @__PURE__ */ import_react2.default.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:0ms]" }), /* @__PURE__ */ import_react2.default.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:150ms]" }), /* @__PURE__ */ import_react2.default.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:300ms]" })) : /* @__PURE__ */ import_react2.default.createElement("div", { className: `prose prose-sm max-w-none ${isUser ? "prose-invert" : "dark:prose-invert"}` }, /* @__PURE__ */ import_react2.default.createElement(import_react_markdown.default, { remarkPlugins: [import_remark_gfm.default] }, message.content), isStreaming && message.content && /* @__PURE__ */ import_react2.default.createElement("span", { className: "inline-block w-1.5 h-3.5 ml-1 bg-emerald-500 animate-pulse align-middle" }))
   ), !isUser && sources && sources.length > 0 && /* @__PURE__ */ import_react2.default.createElement("div", { className: "w-full" }, /* @__PURE__ */ import_react2.default.createElement(
     "button",
     {
@@ -186,7 +186,6 @@ function useConfig() {
 
 // src/hooks/useRagChat.ts
 var import_react4 = require("react");
-var import_axios = __toESM(require("axios"));
 
 // src/hooks/useStoredMessages.ts
 var React4 = __toESM(require("react"));
@@ -240,7 +239,6 @@ function useRagChat(projectId, options = {}) {
   }, [messages]);
   const sendMessage = (0, import_react4.useCallback)(
     async (text, options2) => {
-      var _a, _b, _c;
       const trimmed = text.trim();
       if (!trimmed || isLoading) return;
       lastInputRef.current = trimmed;
@@ -257,22 +255,77 @@ function useRagChat(projectId, options = {}) {
       setIsLoading(true);
       try {
         const history = messagesRef.current.map(({ role, content }) => ({ role, content }));
-        const { data } = await import_axios.default.post(apiUrl, {
-          message: trimmed,
-          history,
-          namespace: namespace != null ? namespace : projectId
+        const response = await fetch(apiUrl, {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({
+            message: trimmed,
+            history,
+            namespace: namespace != null ? namespace : projectId
+          })
         });
-        const assistantMessage = {
-          id: `assistant_${Date.now()}`,
+        if (!response.ok) {
+          const errorData = await response.json().catch(() => ({}));
+          throw new Error(errorData.error || `Server returned ${response.status}`);
+        }
+        if (!response.body) {
+          throw new Error("No response body received from server");
+        }
+        const reader = response.body.getReader();
+        const decoder = new TextDecoder();
+        let assistantContent = "";
+        let sources = [];
+        const assistantMessageId = `assistant_${Date.now()}`;
+        setMessages((prev) => [
+          ...prev,
+          {
+            id: assistantMessageId,
+            role: "assistant",
+            content: "",
+            createdAt: (/* @__PURE__ */ new Date()).toISOString()
+          }
+        ]);
+        while (true) {
+          const { done, value } = await reader.read();
+          if (done) break;
+          const chunk = decoder.decode(value, { stream: true });
+          if (chunk.includes("__METADATA__")) {
+            const [text2, metadataStr] = chunk.split("__METADATA__");
+            if (text2) assistantContent += text2;
+            try {
+              const meta = JSON.parse(metadataStr);
+              sources = meta.sources || [];
+            } catch (e) {
+              console.warn("[useRagChat] Failed to parse metadata");
+            }
+          } else if (chunk.includes("__ERROR__")) {
+            const [text2, errorStr] = chunk.split("__ERROR__");
+            if (text2) assistantContent += text2;
+            try {
+              const err = JSON.parse(errorStr);
+              setError(err.error || "Stream error");
+            } catch (e) {
+              setError("An error occurred during streaming");
+            }
+          } else {
+            assistantContent += chunk;
+          }
+          setMessages(
+            (prev) => prev.map(
+              (msg) => msg.id === assistantMessageId ? __spreadProps(__spreadValues({}, msg), { content: assistantContent, sources: sources.length > 0 ? sources : msg.sources }) : msg
+            )
+          );
+        }
+        const finalAssistantMessage = {
+          id: assistantMessageId,
           role: "assistant",
-          content: data.reply,
-          sources: data.sources,
+          content: assistantContent,
+          sources: sources.length > 0 ? sources : void 0,
           createdAt: (/* @__PURE__ */ new Date()).toISOString()
         };
-        setMessages((prev) => [...prev, assistantMessage]);
-        onReply == null ? void 0 : onReply(assistantMessage);
+        onReply == null ? void 0 : onReply(finalAssistantMessage);
       } catch (err) {
-        const msg = import_axios.default.isAxiosError(err) ? (_c = (_b = (_a = err.response) == null ? void 0 : _a.data) == null ? void 0 : _b.error) != null ? _c : err.message : "Something went wrong. Please try again.";
+        const msg = err instanceof Error ? err.message : "Something went wrong. Please try again.";
         setError(msg);
         onError == null ? void 0 : onError(msg);
       } finally {
@@ -324,6 +377,7 @@ var BORDER_RADIUS_MAP = {
 
 // src/components/ChatWindow.tsx
 function ChatWindow({ className = "", style, onClose, showClose = false }) {
+  var _a;
   const { ui, projectId } = useConfig();
   const [input, setInput] = (0, import_react6.useState)("");
   const [mounted, setMounted] = (0, import_react6.useState)(false);
@@ -336,9 +390,9 @@ function ChatWindow({ className = "", style, onClose, showClose = false }) {
     setMounted(true);
   }, []);
   (0, import_react6.useEffect)(() => {
-    var _a;
+    var _a2;
     if (messages.length > 0 || isLoading) {
-      (_a = bottomRef.current) == null ? void 0 : _a.scrollIntoView({ behavior: "smooth" });
+      (_a2 = bottomRef.current) == null ? void 0 : _a2.scrollIntoView({ behavior: "smooth" });
     }
   }, [messages, isLoading]);
   const sendMessage = (0, import_react6.useCallback)(async () => {
@@ -347,8 +401,8 @@ function ChatWindow({ className = "", style, onClose, showClose = false }) {
     setInput("");
     await send(text);
     window.setTimeout(() => {
-      var _a;
-      return (_a = inputRef.current) == null ? void 0 : _a.focus();
+      var _a2;
+      return (_a2 = inputRef.current) == null ? void 0 : _a2.focus();
     }, 50);
   }, [input, isLoading, send]);
   const handleKeyDown = (e) => {
@@ -419,25 +473,26 @@ function ChatWindow({ className = "", style, onClose, showClose = false }) {
           {
             key: suggestion,
             onClick: () => {
-              var _a;
+              var _a2;
               setInput(suggestion);
-              (_a = inputRef.current) == null ? void 0 : _a.focus();
+              (_a2 = inputRef.current) == null ? void 0 : _a2.focus();
             },
             className: "px-3 py-1.5 rounded-full text-xs border border-slate-200 dark:border-white/10 text-slate-600 dark:text-white/50 hover:text-slate-900 dark:hover:text-white/80 hover:border-slate-300 dark:hover:border-white/20 hover:bg-slate-100 dark:hover:bg-white/5 transition-all"
           },
           suggestion
         )
       )))
-    ) : messages.map((msg) => /* @__PURE__ */ import_react6.default.createElement(
+    ) : messages.map((msg, index) => /* @__PURE__ */ import_react6.default.createElement(
       MessageBubble,
       {
         key: msg.id,
         message: msg,
         sources: msg.sources,
+        isStreaming: isLoading && index === messages.length - 1 && msg.role === "assistant",
         primaryColor: ui.primaryColor,
         accentColor: ui.accentColor
       }
-    )), isLoading && /* @__PURE__ */ import_react6.default.createElement(
+    )), isLoading && ((_a = messages[messages.length - 1]) == null ? void 0 : _a.role) !== "assistant" && /* @__PURE__ */ import_react6.default.createElement(
       MessageBubble,
       {
         message: { role: "assistant", content: "" },

package/dist/index.mjs

@@ -73,7 +73,7 @@ function MessageBubble({
       className: `relative px-4 py-3 rounded-2xl text-sm leading-relaxed shadow-sm dark:shadow-lg ${isUser ? "text-white rounded-tr-sm" : "bg-white dark:bg-white/5 border border-slate-200 dark:border-white/10 text-slate-800 dark:text-white/90 rounded-tl-sm"}`,
       style: isUser ? { background: `linear-gradient(135deg, ${primaryColor}, ${accentColor})` } : {}
     },
-    isStreaming ? /* @__PURE__ */ React2.createElement("span", { className: "flex items-center gap-1" }, /* @__PURE__ */ React2.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:0ms]" }), /* @__PURE__ */ React2.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:150ms]" }), /* @__PURE__ */ React2.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:300ms]" })) : /* @__PURE__ */ React2.createElement("div", { className: `prose prose-sm max-w-none ${isUser ? "prose-invert" : "dark:prose-invert"}` }, /* @__PURE__ */ React2.createElement(ReactMarkdown, { remarkPlugins: [remarkGfm] }, message.content))
+    isStreaming && !message.content ? /* @__PURE__ */ React2.createElement("span", { className: "flex items-center gap-1 py-1" }, /* @__PURE__ */ React2.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:0ms]" }), /* @__PURE__ */ React2.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:150ms]" }), /* @__PURE__ */ React2.createElement("span", { className: "w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:300ms]" })) : /* @__PURE__ */ React2.createElement("div", { className: `prose prose-sm max-w-none ${isUser ? "prose-invert" : "dark:prose-invert"}` }, /* @__PURE__ */ React2.createElement(ReactMarkdown, { remarkPlugins: [remarkGfm] }, message.content), isStreaming && message.content && /* @__PURE__ */ React2.createElement("span", { className: "inline-block w-1.5 h-3.5 ml-1 bg-emerald-500 animate-pulse align-middle" }))
   ), !isUser && sources && sources.length > 0 && /* @__PURE__ */ React2.createElement("div", { className: "w-full" }, /* @__PURE__ */ React2.createElement(
     "button",
     {
@@ -125,7 +125,6 @@ function useConfig() {
 
 // src/hooks/useRagChat.ts
 import { useState as useState2, useCallback, useRef, useEffect as useEffect2 } from "react";
-import axios from "axios";
 
 // src/hooks/useStoredMessages.ts
 import * as React4 from "react";
@@ -179,7 +178,6 @@ function useRagChat(projectId, options = {}) {
   }, [messages]);
   const sendMessage = useCallback(
     async (text, options2) => {
-      var _a, _b, _c;
       const trimmed = text.trim();
       if (!trimmed || isLoading) return;
       lastInputRef.current = trimmed;
@@ -196,22 +194,77 @@ function useRagChat(projectId, options = {}) {
       setIsLoading(true);
       try {
         const history = messagesRef.current.map(({ role, content }) => ({ role, content }));
-        const { data } = await axios.post(apiUrl, {
-          message: trimmed,
-          history,
-          namespace: namespace != null ? namespace : projectId
+        const response = await fetch(apiUrl, {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({
+            message: trimmed,
+            history,
+            namespace: namespace != null ? namespace : projectId
+          })
         });
-        const assistantMessage = {
-          id: `assistant_${Date.now()}`,
+        if (!response.ok) {
+          const errorData = await response.json().catch(() => ({}));
+          throw new Error(errorData.error || `Server returned ${response.status}`);
+        }
+        if (!response.body) {
+          throw new Error("No response body received from server");
+        }
+        const reader = response.body.getReader();
+        const decoder = new TextDecoder();
+        let assistantContent = "";
+        let sources = [];
+        const assistantMessageId = `assistant_${Date.now()}`;
+        setMessages((prev) => [
+          ...prev,
+          {
+            id: assistantMessageId,
+            role: "assistant",
+            content: "",
+            createdAt: (/* @__PURE__ */ new Date()).toISOString()
+          }
+        ]);
+        while (true) {
+          const { done, value } = await reader.read();
+          if (done) break;
+          const chunk = decoder.decode(value, { stream: true });
+          if (chunk.includes("__METADATA__")) {
+            const [text2, metadataStr] = chunk.split("__METADATA__");
+            if (text2) assistantContent += text2;
+            try {
+              const meta = JSON.parse(metadataStr);
+              sources = meta.sources || [];
+            } catch (e) {
+              console.warn("[useRagChat] Failed to parse metadata");
+            }
+          } else if (chunk.includes("__ERROR__")) {
+            const [text2, errorStr] = chunk.split("__ERROR__");
+            if (text2) assistantContent += text2;
+            try {
+              const err = JSON.parse(errorStr);
+              setError(err.error || "Stream error");
+            } catch (e) {
+              setError("An error occurred during streaming");
+            }
+          } else {
+            assistantContent += chunk;
+          }
+          setMessages(
+            (prev) => prev.map(
+              (msg) => msg.id === assistantMessageId ? __spreadProps(__spreadValues({}, msg), { content: assistantContent, sources: sources.length > 0 ? sources : msg.sources }) : msg
+            )
+          );
+        }
+        const finalAssistantMessage = {
+          id: assistantMessageId,
           role: "assistant",
-          content: data.reply,
-          sources: data.sources,
+          content: assistantContent,
+          sources: sources.length > 0 ? sources : void 0,
           createdAt: (/* @__PURE__ */ new Date()).toISOString()
         };
-        setMessages((prev) => [...prev, assistantMessage]);
-        onReply == null ? void 0 : onReply(assistantMessage);
+        onReply == null ? void 0 : onReply(finalAssistantMessage);
       } catch (err) {
-        const msg = axios.isAxiosError(err) ? (_c = (_b = (_a = err.response) == null ? void 0 : _a.data) == null ? void 0 : _b.error) != null ? _c : err.message : "Something went wrong. Please try again.";
+        const msg = err instanceof Error ? err.message : "Something went wrong. Please try again.";
         setError(msg);
         onError == null ? void 0 : onError(msg);
       } finally {
@@ -281,6 +334,7 @@ var BORDER_RADIUS_MAP = {
 
 // src/components/ChatWindow.tsx
 function ChatWindow({ className = "", style, onClose, showClose = false }) {
+  var _a;
   const { ui, projectId } = useConfig();
   const [input, setInput] = useState3("");
   const [mounted, setMounted] = useState3(false);
@@ -293,9 +347,9 @@ function ChatWindow({ className = "", style, onClose, showClose = false }) {
     setMounted(true);
   }, []);
   useEffect3(() => {
-    var _a;
+    var _a2;
     if (messages.length > 0 || isLoading) {
-      (_a = bottomRef.current) == null ? void 0 : _a.scrollIntoView({ behavior: "smooth" });
+      (_a2 = bottomRef.current) == null ? void 0 : _a2.scrollIntoView({ behavior: "smooth" });
     }
   }, [messages, isLoading]);
   const sendMessage = useCallback2(async () => {
@@ -304,8 +358,8 @@ function ChatWindow({ className = "", style, onClose, showClose = false }) {
     setInput("");
     await send(text);
     window.setTimeout(() => {
-      var _a;
-      return (_a = inputRef.current) == null ? void 0 : _a.focus();
+      var _a2;
+      return (_a2 = inputRef.current) == null ? void 0 : _a2.focus();
     }, 50);
   }, [input, isLoading, send]);
   const handleKeyDown = (e) => {
@@ -376,25 +430,26 @@ function ChatWindow({ className = "", style, onClose, showClose = false }) {
           {
             key: suggestion,
             onClick: () => {
-              var _a;
+              var _a2;
               setInput(suggestion);
-              (_a = inputRef.current) == null ? void 0 : _a.focus();
+              (_a2 = inputRef.current) == null ? void 0 : _a2.focus();
             },
             className: "px-3 py-1.5 rounded-full text-xs border border-slate-200 dark:border-white/10 text-slate-600 dark:text-white/50 hover:text-slate-900 dark:hover:text-white/80 hover:border-slate-300 dark:hover:border-white/20 hover:bg-slate-100 dark:hover:bg-white/5 transition-all"
           },
           suggestion
         )
       )))
-    ) : messages.map((msg) => /* @__PURE__ */ React6.createElement(
+    ) : messages.map((msg, index) => /* @__PURE__ */ React6.createElement(
       MessageBubble,
       {
         key: msg.id,
         message: msg,
         sources: msg.sources,
+        isStreaming: isLoading && index === messages.length - 1 && msg.role === "assistant",
         primaryColor: ui.primaryColor,
         accentColor: ui.accentColor
       }
-    )), isLoading && /* @__PURE__ */ React6.createElement(
+    )), isLoading && ((_a = messages[messages.length - 1]) == null ? void 0 : _a.role) !== "assistant" && /* @__PURE__ */ React6.createElement(
       MessageBubble,
       {
         message: { role: "assistant", content: "" },

package/dist/server.d.mts

@@ -111,26 +111,7 @@ declare class ProviderHealthCheck {
 declare class VectorPlugin {
     private config;
     private pipeline;
-    /**
-     * Initializes the plugin with the host configuration.
-     * @param hostConfig - Configuration object passed from the host application.
-     * @throws Error if configuration is invalid or providers are unhealthy
-     *
-     * @example
-     * const plugin = new VectorPlugin({
-     *   projectId: 'my-app',
-     *   vectorDb: {
-     *     provider: 'pinecone',
-     *     indexName: 'my-index',
-     *     options: { apiKey: process.env.PINECONE_API_KEY }
-     *   },
-     *   llm: {
-     *     provider: 'openai',
-     *     model: 'gpt-4o',
-     *     apiKey: process.env.OPENAI_API_KEY
-     *   }
-     * });
-     */
+    private validationPromise;
     constructor(hostConfig?: Partial<RagConfig>);
     /**
      * Get the current resolved configuration.

package/dist/server.d.ts

@@ -111,26 +111,7 @@ declare class ProviderHealthCheck {
 declare class VectorPlugin {
     private config;
     private pipeline;
-    /**
-     * Initializes the plugin with the host configuration.
-     * @param hostConfig - Configuration object passed from the host application.
-     * @throws Error if configuration is invalid or providers are unhealthy
-     *
-     * @example
-     * const plugin = new VectorPlugin({
-     *   projectId: 'my-app',
-     *   vectorDb: {
-     *     provider: 'pinecone',
-     *     indexName: 'my-index',
-     *     options: { apiKey: process.env.PINECONE_API_KEY }
-     *   },
-     *   llm: {
-     *     provider: 'openai',
-     *     model: 'gpt-4o',
-     *     apiKey: process.env.OPENAI_API_KEY
-     *   }
-     * });
-     */
+    private validationPromise;
     constructor(hostConfig?: Partial<RagConfig>);
     /**
      * Get the current resolved configuration.

package/dist/server.js

@@ -558,7 +558,7 @@ var init_MongoDBProvider = __esm({
         if (!opts.uri || !opts.database || !opts.collection) {
           throw new Error("[MongoDBProvider] options uri, database, and collection are required.");
         }
-        this.client = new import_mongodb.MongoClient(opts.uri);
+        this.client = new import_mongodb.MongoClient(opts.uri, { serverSelectionTimeoutMS: 5e3 });
         this.dbName = opts.database;
         this.collectionName = opts.collection;
         this.embeddingKey = opts.embeddingKey || "embedding";
@@ -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: [] };
     }
   }
@@ -3522,6 +3535,7 @@ var Pipeline = class {
   async initialize() {
     var _a;
     if (this.initialised) return;
+    console.log(`[Pipeline] Initializing with provider: ${this.config.vectorDb.provider}...`);
     this.vectorDB = await ProviderRegistry.createVectorProvider(this.config.vectorDb);
     const { llmProvider, embeddingProvider } = await EmbeddingStrategyResolver.resolve(
       this.config.llm,
@@ -3906,29 +3920,9 @@ var ProviderHealthCheck = class {
 
 // src/core/VectorPlugin.ts
 var VectorPlugin = class {
-  /**
-   * Initializes the plugin with the host configuration.
-   * @param hostConfig - Configuration object passed from the host application.
-   * @throws Error if configuration is invalid or providers are unhealthy
-   *
-   * @example
-   * const plugin = new VectorPlugin({
-   *   projectId: 'my-app',
-   *   vectorDb: {
-   *     provider: 'pinecone',
-   *     indexName: 'my-index',
-   *     options: { apiKey: process.env.PINECONE_API_KEY }
-   *   },
-   *   llm: {
-   *     provider: 'openai',
-   *     model: 'gpt-4o',
-   *     apiKey: process.env.OPENAI_API_KEY
-   *   }
-   * });
-   */
   constructor(hostConfig) {
     this.config = ConfigResolver.resolve(hostConfig);
-    ConfigValidator.validateAndThrow(this.config);
+    this.validationPromise = ConfigValidator.validateAndThrow(this.config);
     this.pipeline = new Pipeline(this.config);
   }
   /**
@@ -3954,6 +3948,7 @@ var VectorPlugin = class {
    * Run a chat query.
    */
   async chat(message, history = [], namespace) {
+    await this.validationPromise;
     return this.pipeline.ask(message, history, namespace);
   }
   /**
@@ -3961,6 +3956,7 @@ var VectorPlugin = class {
    */
   chatStream(_0) {
     return __asyncGenerator(this, arguments, function* (message, history = [], namespace) {
+      yield new __await(this.validationPromise);
       yield* __yieldStar(this.pipeline.askStream(message, history, namespace));
     });
   }
@@ -3968,6 +3964,7 @@ var VectorPlugin = class {
    * Ingest documents into the vector database.
    */
   async ingest(documents, namespace) {
+    await this.validationPromise;
     return this.pipeline.ingest(documents, namespace);
   }
 };

package/dist/server.mjs

@@ -34,7 +34,7 @@ import {
   createIngestHandler,
   createUploadHandler,
   getRagConfig
-} from "./chunk-5W2YWFT3.mjs";
+} from "./chunk-3JR3SAWX.mjs";
 import "./chunk-YLTMFW4M.mjs";
 import {
   PineconeProvider
@@ -44,7 +44,7 @@ import {
 } from "./chunk-FLOSGE6A.mjs";
 import {
   MongoDBProvider
-} from "./chunk-CD6TSNL4.mjs";
+} from "./chunk-73I6VWU3.mjs";
 import {
   MilvusProvider
 } from "./chunk-U55XRW3U.mjs";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@retrivora-ai/rag-engine",
-  "version": "1.0.1",
+  "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/components/ChatWindow.tsx

@@ -170,19 +170,20 @@ export function ChatWindow({ className = '', style, onClose, showClose = false }
             </div>
           </div>
         ) : (
-          messages.map((msg) => (
+          messages.map((msg, index) => (
             <MessageBubble
               key={msg.id}
               message={msg}
               sources={msg.sources}
+              isStreaming={isLoading && index === messages.length - 1 && msg.role === 'assistant'}
               primaryColor={ui.primaryColor}
               accentColor={ui.accentColor}
             />
           ))
         )}
 
-        {/* Typing indicator */}
-        {isLoading && (
+        {/* Typing indicator (only if assistant hasn't started replying yet) */}
+        {isLoading && messages[messages.length - 1]?.role !== 'assistant' && (
           <MessageBubble
             message={{ role: 'assistant', content: '' }}
             isStreaming

package/src/components/MessageBubble.tsx

@@ -55,8 +55,8 @@ export function MessageBubble({
               : {}
           }
         >
-          {isStreaming ? (
-            <span className="flex items-center gap-1">
+          {isStreaming && !message.content ? (
+            <span className="flex items-center gap-1 py-1">
               <span className="w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:0ms]" />
               <span className="w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:150ms]" />
               <span className="w-1.5 h-1.5 rounded-full bg-slate-400 dark:bg-white/60 animate-bounce [animation-delay:300ms]" />
@@ -66,6 +66,9 @@ export function MessageBubble({
               <ReactMarkdown remarkPlugins={[remarkGfm]}>
                 {message.content}
               </ReactMarkdown>
+              {isStreaming && message.content && (
+                <span className="inline-block w-1.5 h-3.5 ml-1 bg-emerald-500 animate-pulse align-middle" />
+              )}
             </div>
           )}
         </div>

package/src/core/Pipeline.ts

@@ -50,6 +50,7 @@ export class Pipeline {
   async initialize(): Promise<void> {
     if (this.initialised) return;
 
+    console.log(`[Pipeline] Initializing with provider: ${this.config.vectorDb.provider}...`);
     // Resolve vector DB provider
     this.vectorDB = await ProviderRegistry.createVectorProvider(this.config.vectorDb);
 

package/src/core/VectorPlugin.ts

@@ -20,32 +20,13 @@ import { ChatMessage } from '../llm/ILLMProvider';
 export class VectorPlugin {
   private config: RagConfig;
   private pipeline: Pipeline;
+  private validationPromise: Promise<void>;
 
-  /**
-   * Initializes the plugin with the host configuration.
-   * @param hostConfig - Configuration object passed from the host application.
-   * @throws Error if configuration is invalid or providers are unhealthy
-   *
-   * @example
-   * const plugin = new VectorPlugin({
-   *   projectId: 'my-app',
-   *   vectorDb: {
-   *     provider: 'pinecone',
-   *     indexName: 'my-index',
-   *     options: { apiKey: process.env.PINECONE_API_KEY }
-   *   },
-   *   llm: {
-   *     provider: 'openai',
-   *     model: 'gpt-4o',
-   *     apiKey: process.env.OPENAI_API_KEY
-   *   }
-   * });
-   */
   constructor(hostConfig?: Partial<RagConfig>) {
     this.config = ConfigResolver.resolve(hostConfig);
     
-    // Validate configuration early with detailed error messages
-    ConfigValidator.validateAndThrow(this.config);
+    // Start validation early
+    this.validationPromise = ConfigValidator.validateAndThrow(this.config);
     
     this.pipeline = new Pipeline(this.config);
   }
@@ -80,6 +61,7 @@ export class VectorPlugin {
    * Run a chat query.
    */
   async chat(message: string, history: ChatMessage[] = [], namespace?: string): Promise<ChatResponse> {
+    await this.validationPromise;
     return this.pipeline.ask(message, history, namespace);
   }
 
@@ -87,6 +69,7 @@ export class VectorPlugin {
    * Run a streaming chat query.
    */
   async *chatStream(message: string, history: ChatMessage[] = [], namespace?: string) {
+    await this.validationPromise;
     yield* this.pipeline.askStream(message, history, namespace);
   }
 
@@ -94,6 +77,7 @@ export class VectorPlugin {
    * Ingest documents into the vector database.
    */
   async ingest(documents: IngestDocument[], namespace?: string): Promise<Array<{ docId: string | number; chunksIngested: number }>> {
+    await this.validationPromise;
     return this.pipeline.ingest(documents, namespace);
   }
 

package/src/handlers/index.ts

@@ -40,22 +40,30 @@ export function createStreamHandler(config?: Partial<RagConfig>) {
 
   return async function POST(req: NextRequest) {
     try {
-      const { message, history = [], namespace } = await req.json();
+      const body = await req.json();
+      const { message, history = [], namespace } = body;
       
       const encoder = new TextEncoder();
       const stream = new ReadableStream({
         async start(controller) {
-          const pipelineStream = plugin.chatStream(message, history, namespace);
-          
-          for await (const chunk of pipelineStream) {
-            if (typeof chunk === 'string') {
-              controller.enqueue(encoder.encode(chunk));
-            } else {
-              // Yield metadata/sources at the end
-              controller.enqueue(encoder.encode(`\n\n__METADATA__${JSON.stringify(chunk)}`));
+          try {
+            const pipelineStream = plugin.chatStream(message, history, namespace);
+            
+            for await (const chunk of pipelineStream) {
+              if (typeof chunk === 'string') {
+                controller.enqueue(encoder.encode(chunk));
+              } else {
+                // Yield metadata/sources at the end
+                controller.enqueue(encoder.encode(`\n\n__METADATA__${JSON.stringify(chunk)}`));
+              }
             }
+            controller.close();
+          } catch (streamError) {
+            console.error('[createStreamHandler] Stream processing error:', streamError);
+            const errorMessage = streamError instanceof Error ? streamError.message : String(streamError);
+            controller.enqueue(encoder.encode(`\n\n__ERROR__${JSON.stringify({ error: errorMessage })}`));
+            controller.close();
           }
-          controller.close();
         },
       });
 

package/src/hooks/useRagChat.ts

@@ -14,7 +14,6 @@
  */
 
 import { useState, useCallback, useRef, useEffect } from 'react';
-import axios from 'axios';
 import { VectorMatch } from '@/types';
 import { useStoredMessages } from './useStoredMessages';
 
@@ -111,26 +110,91 @@ export function useRagChat(
       try {
         const history = messagesRef.current.map(({ role, content }) => ({ role, content }));
 
-        const { data } = await axios.post(apiUrl, {
-          message: trimmed,
-          history,
-          namespace: namespace ?? projectId,
+        const response = await fetch(apiUrl, {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({
+            message: trimmed,
+            history,
+            namespace: namespace ?? projectId,
+          }),
         });
 
-        const assistantMessage: RagMessage = {
-          id: `assistant_${Date.now()}`,
+        if (!response.ok) {
+          const errorData = await response.json().catch(() => ({}));
+          throw new Error(errorData.error || `Server returned ${response.status}`);
+        }
+
+        if (!response.body) {
+          throw new Error('No response body received from server');
+        }
+
+        const reader = response.body.getReader();
+        const decoder = new TextDecoder();
+        let assistantContent = '';
+        let sources: VectorMatch[] = [];
+
+        // Add a placeholder assistant message that we will update
+        const assistantMessageId = `assistant_${Date.now()}`;
+        setMessages((prev) => [
+          ...prev,
+          {
+            id: assistantMessageId,
+            role: 'assistant',
+            content: '',
+            createdAt: new Date().toISOString(),
+          },
+        ]);
+
+        while (true) {
+          const { done, value } = await reader.read();
+          if (done) break;
+
+          const chunk = decoder.decode(value, { stream: true });
+          
+          // Handle protocol tokens
+          if (chunk.includes('__METADATA__')) {
+            const [text, metadataStr] = chunk.split('__METADATA__');
+            if (text) assistantContent += text;
+            try {
+              const meta = JSON.parse(metadataStr);
+              sources = meta.sources || [];
+            } catch {
+              console.warn('[useRagChat] Failed to parse metadata');
+            }
+          } else if (chunk.includes('__ERROR__')) {
+            const [text, errorStr] = chunk.split('__ERROR__');
+            if (text) assistantContent += text;
+            try {
+              const err = JSON.parse(errorStr);
+              setError(err.error || 'Stream error');
+            } catch {
+              setError('An error occurred during streaming');
+            }
+          } else {
+            assistantContent += chunk;
+          }
+
+          // Update the specific message
+          setMessages((prev) => 
+            prev.map((msg) => 
+              msg.id === assistantMessageId 
+                ? { ...msg, content: assistantContent, sources: sources.length > 0 ? sources : msg.sources } 
+                : msg
+            )
+          );
+        }
+
+        const finalAssistantMessage: RagMessage = {
+          id: assistantMessageId,
           role: 'assistant',
-          content: data.reply,
-          sources: data.sources,
+          content: assistantContent,
+          sources: sources.length > 0 ? sources : undefined,
           createdAt: new Date().toISOString(),
         };
-
-        setMessages((prev) => [...prev, assistantMessage]);
-        onReply?.(assistantMessage);
+        onReply?.(finalAssistantMessage);
       } catch (err) {
-        const msg = axios.isAxiosError(err)
-          ? (err.response?.data?.error ?? err.message)
-          : 'Something went wrong. Please try again.';
+        const msg = err instanceof Error ? err.message : 'Something went wrong. Please try again.';
         setError(msg);
         onError?.(msg);
       } finally {

package/src/providers/vectordb/MongoDBProvider.ts

@@ -25,7 +25,7 @@ export class MongoDBProvider extends BaseVectorProvider {
     if (!opts.uri || !opts.database || !opts.collection) {
       throw new Error('[MongoDBProvider] options uri, database, and collection are required.');
     }
-    this.client = new MongoClient(opts.uri as string);
+    this.client = new MongoClient(opts.uri as string, { serverSelectionTimeoutMS: 5000 });
     this.dbName = opts.database as string;
     this.collectionName = opts.collection as string;
     this.embeddingKey = (opts.embeddingKey as string) || 'embedding';

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: [] };
     }
   }