@retrivora-ai/rag-engine 1.1.0 → 1.1.2

package/dist/{QdrantProvider-WWXFX2XF.mjs → QdrantProvider-RLJTNGPY.mjs}

@@ -1,6 +1,6 @@
 import {
   QdrantProvider
-} from "./chunk-65S4BQL2.mjs";
+} from "./chunk-PSFPZXHX.mjs";
 import "./chunk-IMP6FUCY.mjs";
 import "./chunk-X4TOT24V.mjs";
 export {

package/dist/{RagConfig-DVovvPmd.d.mts → RagConfig-FyMB_UG6.d.mts}

@@ -103,7 +103,7 @@ interface RetrievalResult {
  * This file serves as the single source of truth for all statically supported
  * providers and stylistic options in the Retrivora AI RAG Engine.
  */
-declare const VECTOR_DB_PROVIDERS: readonly ["pinecone", "pgvector", "postgresql", "mongodb", "milvus", "qdrant", "chromadb", "redis", "weaviate", "rest", "universal_rest", "custom"];
+declare const VECTOR_DB_PROVIDERS: readonly ["pinecone", "pgvector", "postgresql", "mongodb", "qdrant", "milvus", "chromadb", "redis", "weaviate", "rest", "universal_rest", "custom"];
 declare const LLM_PROVIDERS: readonly ["openai", "anthropic", "ollama", "gemini", "rest", "universal_rest", "custom"];
 declare const EMBEDDING_PROVIDERS: readonly ["openai", "ollama", "gemini", "rest", "universal_rest", "custom"];
 declare const GRAPH_DB_PROVIDERS: readonly ["neo4j", "memgraph", "simple", "custom"];

package/dist/{RagConfig-DVovvPmd.d.ts → RagConfig-FyMB_UG6.d.ts}

@@ -103,7 +103,7 @@ interface RetrievalResult {
  * This file serves as the single source of truth for all statically supported
  * providers and stylistic options in the Retrivora AI RAG Engine.
  */
-declare const VECTOR_DB_PROVIDERS: readonly ["pinecone", "pgvector", "postgresql", "mongodb", "milvus", "qdrant", "chromadb", "redis", "weaviate", "rest", "universal_rest", "custom"];
+declare const VECTOR_DB_PROVIDERS: readonly ["pinecone", "pgvector", "postgresql", "mongodb", "qdrant", "milvus", "chromadb", "redis", "weaviate", "rest", "universal_rest", "custom"];
 declare const LLM_PROVIDERS: readonly ["openai", "anthropic", "ollama", "gemini", "rest", "universal_rest", "custom"];
 declare const EMBEDDING_PROVIDERS: readonly ["openai", "ollama", "gemini", "rest", "universal_rest", "custom"];
 declare const GRAPH_DB_PROVIDERS: readonly ["neo4j", "memgraph", "simple", "custom"];

package/dist/{chunk-65S4BQL2.mjs → chunk-PSFPZXHX.mjs}

@@ -11,6 +11,7 @@ import crypto from "crypto";
 var QdrantProvider = class extends BaseVectorProvider {
   constructor(config) {
     super(config);
+    this.schemaDiscovered = false;
     const opts = config.options;
     const baseUrl = opts.baseUrl;
     if (!baseUrl) throw new Error("[QdrantProvider] baseUrl is required");
@@ -19,12 +20,15 @@ var QdrantProvider = class extends BaseVectorProvider {
     this.isFlatPayload = !!opts.flatPayload;
     this.http = axios.create({
       baseURL: baseUrl,
+      timeout: 15e3,
+      // 15s timeout for vector DB operations
       headers: __spreadValues({
         "Content-Type": "application/json"
       }, opts.apiKey ? { "api-key": opts.apiKey } : {})
     });
   }
   async initialize() {
+    if (this.schemaDiscovered) return;
     await this.ping();
     await this.ensureCollection();
     console.log(`[QdrantProvider] \u{1F50D} Discovering schema for collection "${this.indexName}"...`);
@@ -34,13 +38,16 @@ var QdrantProvider = class extends BaseVectorProvider {
     const allFields = [.../* @__PURE__ */ new Set([...discoveredFields, ...configFields])];
     if (allFields.length > 0) {
       console.log(`[QdrantProvider] \u{1F6E0} Ensuring indexes for ${allFields.length} discovered fields...`);
-      for (const fieldInfo of allFields) {
-        const [fieldName, schemaType] = fieldInfo.split(":");
-        await this.ensureIndex(fieldName, schemaType || "keyword");
-      }
+      await Promise.all(
+        allFields.map(async (fieldInfo) => {
+          const [fieldName, schemaType] = fieldInfo.split(":");
+          return this.ensureIndex(fieldName, schemaType || "keyword");
+        })
+      );
     } else {
       console.log(`[QdrantProvider] \u2139\uFE0F No fields discovered for indexing.`);
     }
+    this.schemaDiscovered = true;
   }
   /**
    * Samples points from the collection to discover available payload fields.
@@ -177,7 +184,7 @@ var QdrantProvider = class extends BaseVectorProvider {
       const p = res.payload || {};
       let content = p[this.contentField] || "";
       if (!content) {
-        const stringFields = Object.entries(p).filter(([_, v]) => typeof v === "string").map(([k, v]) => ({ key: k, val: v }));
+        const stringFields = Object.entries(p).filter(([, v]) => typeof v === "string").map(([k, v]) => ({ key: k, val: v }));
         if (stringFields.length > 0) {
           const bestMatch = stringFields.sort((a, b) => b.val.length - a.val.length)[0];
           content = bestMatch.val;
@@ -202,7 +209,7 @@ var QdrantProvider = class extends BaseVectorProvider {
     });
     return results;
   }
-  async delete(id, _namespace) {
+  async delete(id) {
     await this.http.post(`/collections/${this.indexName}/points/delete`, {
       points: [this.normalizeId(id)]
     });

package/dist/{chunk-G2LVK46T.mjs → chunk-XEWWAHCM.mjs}

@@ -21,8 +21,8 @@ var VECTOR_DB_PROVIDERS = [
   "pgvector",
   "postgresql",
   "mongodb",
-  "milvus",
   "qdrant",
+  "milvus",
   "chromadb",
   "redis",
   "weaviate",
@@ -333,6 +333,56 @@ ${context}`
     });
     return (_h = (_g = (_f = completion.choices[0]) == null ? void 0 : _f.message) == null ? void 0 : _g.content) != null ? _h : "";
   }
+  chatStream(messages, context, options) {
+    return __asyncGenerator(this, null, function* () {
+      var _a, _b, _c, _d, _e, _f, _g;
+      const systemContent = (_a = this.llmConfig.systemPrompt) != null ? _a : `You are a helpful assistant. Answer questions based on the provided context.
+
+Context:
+${context}`;
+      const systemMessage = {
+        role: "system",
+        content: systemContent.includes("{{context}}") ? systemContent.replace("{{context}}", context) : `${systemContent}
+
+Context:
+${context}`
+      };
+      const formattedMessages = [
+        systemMessage,
+        ...messages.map((m) => ({
+          role: m.role,
+          content: m.content
+        }))
+      ];
+      const stream = yield new __await(this.client.chat.completions.create({
+        model: this.llmConfig.model,
+        messages: formattedMessages,
+        max_tokens: (_c = (_b = options == null ? void 0 : options.maxTokens) != null ? _b : this.llmConfig.maxTokens) != null ? _c : 1024,
+        temperature: (_e = (_d = options == null ? void 0 : options.temperature) != null ? _d : this.llmConfig.temperature) != null ? _e : 0.7,
+        stop: options == null ? void 0 : options.stop,
+        stream: true
+      }));
+      if (!stream) {
+        throw new Error("[OpenAIProvider] completions.create stream is undefined");
+      }
+      try {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+          const chunk = temp.value;
+          const content = ((_g = (_f = chunk.choices[0]) == null ? void 0 : _f.delta) == null ? void 0 : _g.content) || "";
+          if (content) yield content;
+        }
+      } catch (temp) {
+        error = [temp];
+      } finally {
+        try {
+          more && (temp = iter.return) && (yield new __await(temp.call(iter)));
+        } finally {
+          if (error)
+            throw error[0];
+        }
+      }
+    });
+  }
   async embed(text, options) {
     const results = await this.batchEmbed([text], options);
     return results[0];
@@ -438,6 +488,50 @@ ${context}`;
     const block = response.content[0];
     return block.type === "text" ? block.text : "";
   }
+  chatStream(messages, context, options) {
+    return __asyncGenerator(this, null, function* () {
+      var _a, _b, _c;
+      const systemPrompt = (_a = this.llmConfig.systemPrompt) != null ? _a : `You are a helpful assistant. Use the context below to answer the user's question accurately.
+
+Context:
+${context}`;
+      const system = systemPrompt.includes("{{context}}") ? systemPrompt.replace("{{context}}", context) : `${systemPrompt}
+
+Context:
+${context}`;
+      const anthropicMessages = messages.map((m) => ({
+        role: m.role === "assistant" ? "assistant" : "user",
+        content: m.content
+      }));
+      const stream = yield new __await(this.client.messages.create({
+        model: this.llmConfig.model,
+        max_tokens: (_c = (_b = options == null ? void 0 : options.maxTokens) != null ? _b : this.llmConfig.maxTokens) != null ? _c : 1024,
+        system,
+        messages: anthropicMessages,
+        stream: true
+      }));
+      if (!stream) {
+        throw new Error("[AnthropicProvider] messages.create stream is undefined");
+      }
+      try {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+          const chunk = temp.value;
+          if (chunk.type === "content_block_delta" && chunk.delta.type === "text_delta") {
+            yield chunk.delta.text;
+          }
+        }
+      } catch (temp) {
+        error = [temp];
+      } finally {
+        try {
+          more && (temp = iter.return) && (yield new __await(temp.call(iter)));
+        } finally {
+          if (error)
+            throw error[0];
+        }
+      }
+    });
+  }
   async embed(text, options) {
     void text;
     void options;
@@ -543,7 +637,7 @@ var OllamaProvider = class {
   }
   chatStream(messages, context, options) {
     return __asyncGenerator(this, null, function* () {
-      var _a, _b, _c, _d, _e;
+      var _a, _b, _c, _d, _e, _f;
       const system = this.buildSystemPrompt(context);
       const response = yield new __await(this.http.post("/api/chat", {
         model: this.llmConfig.model,
@@ -557,11 +651,19 @@ var OllamaProvider = class {
           ...messages.map((m) => ({ role: m.role, content: m.content }))
         ]
       }, { responseType: "stream" }));
+      let lineBuffer = "";
+      const stream = response.data;
+      if (!stream) {
+        throw new Error("[OllamaProvider] response.data is undefined for stream request. Axios might not support streams in this environment.");
+      }
       try {
-        for (var iter = __forAwait(response.data), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
           const chunk = temp.value;
-          const lines = chunk.toString().split("\n").filter(Boolean);
+          lineBuffer += chunk.toString();
+          const lines = lineBuffer.split("\n");
+          lineBuffer = lines.pop() || "";
           for (const line of lines) {
+            if (!line.trim()) continue;
             try {
               const json = JSON.parse(line);
               if ((_e = json.message) == null ? void 0 : _e.content) {
@@ -569,6 +671,7 @@ var OllamaProvider = class {
               }
               if (json.done) return;
             } catch (e) {
+              console.warn("[OllamaProvider] Failed to parse streaming line:", line);
             }
           }
         }
@@ -582,6 +685,13 @@ var OllamaProvider = class {
             throw error[0];
         }
       }
+      if (lineBuffer.trim()) {
+        try {
+          const json = JSON.parse(lineBuffer);
+          if ((_f = json.message) == null ? void 0 : _f.content) yield json.message.content;
+        } catch (e) {
+        }
+      }
     });
   }
   buildSystemPrompt(context) {
@@ -727,6 +837,53 @@ ${context}`;
     });
     return (_f = response.text) != null ? _f : "";
   }
+  chatStream(messages, context, options) {
+    return __asyncGenerator(this, null, function* () {
+      var _a, _b, _c, _d, _e, _f;
+      const systemPrompt = (_a = this.llmConfig.systemPrompt) != null ? _a : `You are a helpful assistant. Answer questions based on the provided context.
+
+Context:
+${context}`;
+      const systemInstruction = systemPrompt.includes("{{context}}") ? systemPrompt.replace("{{context}}", context) : `${systemPrompt}
+
+Context:
+${context}`;
+      const geminiMessages = messages.map((m) => ({
+        role: m.role === "assistant" ? "model" : "user",
+        parts: [{ text: m.content }]
+      }));
+      const result = yield new __await(this.client.models.generateContentStream({
+        model: this.llmConfig.model,
+        contents: geminiMessages,
+        config: {
+          systemInstruction,
+          temperature: (_c = (_b = options == null ? void 0 : options.temperature) != null ? _b : this.llmConfig.temperature) != null ? _c : 0.7,
+          maxOutputTokens: (_e = (_d = options == null ? void 0 : options.maxTokens) != null ? _d : this.llmConfig.maxTokens) != null ? _e : 1024,
+          stopSequences: options == null ? void 0 : options.stop
+        }
+      }));
+      const stream = (_f = result == null ? void 0 : result.stream) != null ? _f : result;
+      if (!stream) {
+        throw new Error("[GeminiProvider] generateContentStream returned undefined");
+      }
+      try {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+          const chunk = temp.value;
+          const text = chunk.text;
+          if (text) yield text;
+        }
+      } catch (temp) {
+        error = [temp];
+      } finally {
+        try {
+          more && (temp = iter.return) && (yield new __await(temp.call(iter)));
+        } finally {
+          if (error)
+            throw error[0];
+        }
+      }
+    });
+  }
   async embed(text, options) {
     var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k;
     const model = this.sanitizeModel(
@@ -1106,7 +1263,7 @@ var ProviderRegistry = class {
         return MilvusProvider;
       }
       case "qdrant": {
-        const { QdrantProvider } = await import("./QdrantProvider-WWXFX2XF.mjs");
+        const { QdrantProvider } = await import("./QdrantProvider-RLJTNGPY.mjs");
         return QdrantProvider;
       }
       case "chromadb": {
@@ -2299,8 +2456,12 @@ ${context}`;
         }
         const messages = [...history, { role: "user", content: question }];
         if (this.llmProvider.chatStream) {
+          const stream = this.llmProvider.chatStream(messages, context);
+          if (!stream) {
+            throw new Error(`[Pipeline] ${this.config.llm.provider} chatStream returned undefined`);
+          }
           try {
-            for (var iter = __forAwait(this.llmProvider.chatStream(messages, context)), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+            for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
               const chunk = temp.value;
               yield chunk;
             }

package/dist/handlers/index.d.mts

@@ -1,3 +1,3 @@
-import '../RagConfig-DVovvPmd.mjs';
-export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from '../index-D0_2f-43.mjs';
+import '../RagConfig-FyMB_UG6.mjs';
+export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from '../index-D-lfcqlL.mjs';
 import 'next/server';

package/dist/handlers/index.d.ts

@@ -1,3 +1,3 @@
-import '../RagConfig-DVovvPmd.js';
-export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from '../index-CQ0zQ7Zk.js';
+import '../RagConfig-FyMB_UG6.js';
+export { c as createChatHandler, d as createHealthHandler, e as createIngestHandler, f as createStreamHandler, g as createUploadHandler, s as sseErrorFrame, h as sseFrame, i as sseMetaFrame, j as sseTextFrame } from '../index-CYgr00ot.js';
 import 'next/server';

package/dist/handlers/index.js

@@ -842,6 +842,7 @@ var init_QdrantProvider = __esm({
     QdrantProvider = class extends BaseVectorProvider {
       constructor(config) {
         super(config);
+        this.schemaDiscovered = false;
         const opts = config.options;
         const baseUrl = opts.baseUrl;
         if (!baseUrl) throw new Error("[QdrantProvider] baseUrl is required");
@@ -850,12 +851,15 @@ var init_QdrantProvider = __esm({
         this.isFlatPayload = !!opts.flatPayload;
         this.http = import_axios4.default.create({
           baseURL: baseUrl,
+          timeout: 15e3,
+          // 15s timeout for vector DB operations
           headers: __spreadValues({
             "Content-Type": "application/json"
           }, opts.apiKey ? { "api-key": opts.apiKey } : {})
         });
       }
       async initialize() {
+        if (this.schemaDiscovered) return;
         await this.ping();
         await this.ensureCollection();
         console.log(`[QdrantProvider] \u{1F50D} Discovering schema for collection "${this.indexName}"...`);
@@ -865,13 +869,16 @@ var init_QdrantProvider = __esm({
         const allFields = [.../* @__PURE__ */ new Set([...discoveredFields, ...configFields])];
         if (allFields.length > 0) {
           console.log(`[QdrantProvider] \u{1F6E0} Ensuring indexes for ${allFields.length} discovered fields...`);
-          for (const fieldInfo of allFields) {
-            const [fieldName, schemaType] = fieldInfo.split(":");
-            await this.ensureIndex(fieldName, schemaType || "keyword");
-          }
+          await Promise.all(
+            allFields.map(async (fieldInfo) => {
+              const [fieldName, schemaType] = fieldInfo.split(":");
+              return this.ensureIndex(fieldName, schemaType || "keyword");
+            })
+          );
         } else {
           console.log(`[QdrantProvider] \u2139\uFE0F No fields discovered for indexing.`);
         }
+        this.schemaDiscovered = true;
       }
       /**
        * Samples points from the collection to discover available payload fields.
@@ -1008,7 +1015,7 @@ var init_QdrantProvider = __esm({
           const p = res.payload || {};
           let content = p[this.contentField] || "";
           if (!content) {
-            const stringFields = Object.entries(p).filter(([_, v]) => typeof v === "string").map(([k, v]) => ({ key: k, val: v }));
+            const stringFields = Object.entries(p).filter(([, v]) => typeof v === "string").map(([k, v]) => ({ key: k, val: v }));
             if (stringFields.length > 0) {
               const bestMatch = stringFields.sort((a, b) => b.val.length - a.val.length)[0];
               content = bestMatch.val;
@@ -1033,7 +1040,7 @@ var init_QdrantProvider = __esm({
         });
         return results;
       }
-      async delete(id, _namespace) {
+      async delete(id) {
         await this.http.post(`/collections/${this.indexName}/points/delete`, {
           points: [this.normalizeId(id)]
         });
@@ -1623,8 +1630,8 @@ var VECTOR_DB_PROVIDERS = [
   "pgvector",
   "postgresql",
   "mongodb",
-  "milvus",
   "qdrant",
+  "milvus",
   "chromadb",
   "redis",
   "weaviate",
@@ -1932,6 +1939,56 @@ ${context}`
     });
     return (_h = (_g = (_f = completion.choices[0]) == null ? void 0 : _f.message) == null ? void 0 : _g.content) != null ? _h : "";
   }
+  chatStream(messages, context, options) {
+    return __asyncGenerator(this, null, function* () {
+      var _a, _b, _c, _d, _e, _f, _g;
+      const systemContent = (_a = this.llmConfig.systemPrompt) != null ? _a : `You are a helpful assistant. Answer questions based on the provided context.
+
+Context:
+${context}`;
+      const systemMessage = {
+        role: "system",
+        content: systemContent.includes("{{context}}") ? systemContent.replace("{{context}}", context) : `${systemContent}
+
+Context:
+${context}`
+      };
+      const formattedMessages = [
+        systemMessage,
+        ...messages.map((m) => ({
+          role: m.role,
+          content: m.content
+        }))
+      ];
+      const stream = yield new __await(this.client.chat.completions.create({
+        model: this.llmConfig.model,
+        messages: formattedMessages,
+        max_tokens: (_c = (_b = options == null ? void 0 : options.maxTokens) != null ? _b : this.llmConfig.maxTokens) != null ? _c : 1024,
+        temperature: (_e = (_d = options == null ? void 0 : options.temperature) != null ? _d : this.llmConfig.temperature) != null ? _e : 0.7,
+        stop: options == null ? void 0 : options.stop,
+        stream: true
+      }));
+      if (!stream) {
+        throw new Error("[OpenAIProvider] completions.create stream is undefined");
+      }
+      try {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+          const chunk = temp.value;
+          const content = ((_g = (_f = chunk.choices[0]) == null ? void 0 : _f.delta) == null ? void 0 : _g.content) || "";
+          if (content) yield content;
+        }
+      } catch (temp) {
+        error = [temp];
+      } finally {
+        try {
+          more && (temp = iter.return) && (yield new __await(temp.call(iter)));
+        } finally {
+          if (error)
+            throw error[0];
+        }
+      }
+    });
+  }
   async embed(text, options) {
     const results = await this.batchEmbed([text], options);
     return results[0];
@@ -2037,6 +2094,50 @@ ${context}`;
     const block = response.content[0];
     return block.type === "text" ? block.text : "";
   }
+  chatStream(messages, context, options) {
+    return __asyncGenerator(this, null, function* () {
+      var _a, _b, _c;
+      const systemPrompt = (_a = this.llmConfig.systemPrompt) != null ? _a : `You are a helpful assistant. Use the context below to answer the user's question accurately.
+
+Context:
+${context}`;
+      const system = systemPrompt.includes("{{context}}") ? systemPrompt.replace("{{context}}", context) : `${systemPrompt}
+
+Context:
+${context}`;
+      const anthropicMessages = messages.map((m) => ({
+        role: m.role === "assistant" ? "assistant" : "user",
+        content: m.content
+      }));
+      const stream = yield new __await(this.client.messages.create({
+        model: this.llmConfig.model,
+        max_tokens: (_c = (_b = options == null ? void 0 : options.maxTokens) != null ? _b : this.llmConfig.maxTokens) != null ? _c : 1024,
+        system,
+        messages: anthropicMessages,
+        stream: true
+      }));
+      if (!stream) {
+        throw new Error("[AnthropicProvider] messages.create stream is undefined");
+      }
+      try {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+          const chunk = temp.value;
+          if (chunk.type === "content_block_delta" && chunk.delta.type === "text_delta") {
+            yield chunk.delta.text;
+          }
+        }
+      } catch (temp) {
+        error = [temp];
+      } finally {
+        try {
+          more && (temp = iter.return) && (yield new __await(temp.call(iter)));
+        } finally {
+          if (error)
+            throw error[0];
+        }
+      }
+    });
+  }
   async embed(text, options) {
     void text;
     void options;
@@ -2142,7 +2243,7 @@ var OllamaProvider = class {
   }
   chatStream(messages, context, options) {
     return __asyncGenerator(this, null, function* () {
-      var _a, _b, _c, _d, _e;
+      var _a, _b, _c, _d, _e, _f;
       const system = this.buildSystemPrompt(context);
       const response = yield new __await(this.http.post("/api/chat", {
         model: this.llmConfig.model,
@@ -2156,11 +2257,19 @@ var OllamaProvider = class {
           ...messages.map((m) => ({ role: m.role, content: m.content }))
         ]
       }, { responseType: "stream" }));
+      let lineBuffer = "";
+      const stream = response.data;
+      if (!stream) {
+        throw new Error("[OllamaProvider] response.data is undefined for stream request. Axios might not support streams in this environment.");
+      }
       try {
-        for (var iter = __forAwait(response.data), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
           const chunk = temp.value;
-          const lines = chunk.toString().split("\n").filter(Boolean);
+          lineBuffer += chunk.toString();
+          const lines = lineBuffer.split("\n");
+          lineBuffer = lines.pop() || "";
           for (const line of lines) {
+            if (!line.trim()) continue;
             try {
               const json = JSON.parse(line);
               if ((_e = json.message) == null ? void 0 : _e.content) {
@@ -2168,6 +2277,7 @@ var OllamaProvider = class {
               }
               if (json.done) return;
             } catch (e) {
+              console.warn("[OllamaProvider] Failed to parse streaming line:", line);
             }
           }
         }
@@ -2181,6 +2291,13 @@ var OllamaProvider = class {
             throw error[0];
         }
       }
+      if (lineBuffer.trim()) {
+        try {
+          const json = JSON.parse(lineBuffer);
+          if ((_f = json.message) == null ? void 0 : _f.content) yield json.message.content;
+        } catch (e) {
+        }
+      }
     });
   }
   buildSystemPrompt(context) {
@@ -2326,6 +2443,53 @@ ${context}`;
     });
     return (_f = response.text) != null ? _f : "";
   }
+  chatStream(messages, context, options) {
+    return __asyncGenerator(this, null, function* () {
+      var _a, _b, _c, _d, _e, _f;
+      const systemPrompt = (_a = this.llmConfig.systemPrompt) != null ? _a : `You are a helpful assistant. Answer questions based on the provided context.
+
+Context:
+${context}`;
+      const systemInstruction = systemPrompt.includes("{{context}}") ? systemPrompt.replace("{{context}}", context) : `${systemPrompt}
+
+Context:
+${context}`;
+      const geminiMessages = messages.map((m) => ({
+        role: m.role === "assistant" ? "model" : "user",
+        parts: [{ text: m.content }]
+      }));
+      const result = yield new __await(this.client.models.generateContentStream({
+        model: this.llmConfig.model,
+        contents: geminiMessages,
+        config: {
+          systemInstruction,
+          temperature: (_c = (_b = options == null ? void 0 : options.temperature) != null ? _b : this.llmConfig.temperature) != null ? _c : 0.7,
+          maxOutputTokens: (_e = (_d = options == null ? void 0 : options.maxTokens) != null ? _d : this.llmConfig.maxTokens) != null ? _e : 1024,
+          stopSequences: options == null ? void 0 : options.stop
+        }
+      }));
+      const stream = (_f = result == null ? void 0 : result.stream) != null ? _f : result;
+      if (!stream) {
+        throw new Error("[GeminiProvider] generateContentStream returned undefined");
+      }
+      try {
+        for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+          const chunk = temp.value;
+          const text = chunk.text;
+          if (text) yield text;
+        }
+      } catch (temp) {
+        error = [temp];
+      } finally {
+        try {
+          more && (temp = iter.return) && (yield new __await(temp.call(iter)));
+        } finally {
+          if (error)
+            throw error[0];
+        }
+      }
+    });
+  }
   async embed(text, options) {
     var _a, _b, _c, _d, _e, _f, _g, _h, _i, _j, _k;
     const model = this.sanitizeModel(
@@ -3854,8 +4018,12 @@ ${context}`;
         }
         const messages = [...history, { role: "user", content: question }];
         if (this.llmProvider.chatStream) {
+          const stream = this.llmProvider.chatStream(messages, context);
+          if (!stream) {
+            throw new Error(`[Pipeline] ${this.config.llm.provider} chatStream returned undefined`);
+          }
           try {
-            for (var iter = __forAwait(this.llmProvider.chatStream(messages, context)), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
+            for (var iter = __forAwait(stream), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
               const chunk = temp.value;
               yield chunk;
             }

package/dist/handlers/index.mjs

@@ -8,7 +8,7 @@ import {
   sseFrame,
   sseMetaFrame,
   sseTextFrame
-} from "../chunk-G2LVK46T.mjs";
+} from "../chunk-XEWWAHCM.mjs";
 import "../chunk-YLTMFW4M.mjs";
 import "../chunk-X4TOT24V.mjs";
 export {

package/dist/{index-CQ0zQ7Zk.d.ts → index-CYgr00ot.d.ts}

@@ -1,4 +1,4 @@
-import { g as RagConfig, C as ChatMessage, b as ChatResponse, e as IngestDocument } from './RagConfig-DVovvPmd.js';
+import { g as RagConfig, C as ChatMessage, b as ChatResponse, e as IngestDocument } from './RagConfig-FyMB_UG6.js';
 import { NextRequest, NextResponse } from 'next/server';
 
 interface ValidationError {

package/dist/{index-D0_2f-43.d.mts → index-D-lfcqlL.d.mts}

@@ -1,4 +1,4 @@
-import { g as RagConfig, C as ChatMessage, b as ChatResponse, e as IngestDocument } from './RagConfig-DVovvPmd.mjs';
+import { g as RagConfig, C as ChatMessage, b as ChatResponse, e as IngestDocument } from './RagConfig-FyMB_UG6.mjs';
 import { NextRequest, NextResponse } from 'next/server';
 
 interface ValidationError {