@retrivora-ai/rag-engine 2.0.8 → 2.1.2

package/dist/server.js

@@ -594,6 +594,7 @@ function cleanApiKeyOverride(apiKeyOverride) {
 async function dispatchChatCompletion(req, apiKeyOverride) {
   const effectiveKey = cleanApiKeyOverride(apiKeyOverride);
   const provider = resolveProvider(req.model);
+  console.log(`[LLM Gateway Router] Dispatching chat request: model=${req.model}, provider=${provider}, hasGroqKey=${Boolean(process.env.GROQ_API_KEY)}, hasGeminiKey=${Boolean(process.env.GROQ_API_KEY || process.env.GEMINI_API_KEY)}, hasHFToken=${Boolean(process.env.HF_TOKEN)}`);
   try {
     switch (provider) {
       case "groq":
@@ -612,6 +613,10 @@ async function dispatchChatCompletion(req, apiKeyOverride) {
         throw new Error(`Unsupported LLM provider for model: ${req.model}`);
     }
   } catch (error) {
+    console.error(`[LLM Gateway Router] Provider '${provider}' failed for model '${req.model}':`, {
+      message: error.message,
+      stack: error.stack
+    });
     if (/rate[- ]?limit|429|exhausted/i.test(error.message)) {
       const isGroqKey = process.env.GROQ_API_KEY || effectiveKey && effectiveKey.startsWith("gsk_");
       const reqModelLower = req.model.toLowerCase();
@@ -619,7 +624,8 @@ async function dispatchChatCompletion(req, apiKeyOverride) {
         console.warn(`[LLM Gateway] Provider ${provider} rate limited (${error.message}). Falling back to secondary Groq model (llama-3.3-70b-versatile)...`);
         try {
           return await handleGroqRequest(__spreadProps(__spreadValues({}, req), { model: "llama-3.3-70b-versatile" }), effectiveKey);
-        } catch (e) {
+        } catch (fErr) {
+          console.error("[LLM Gateway] Fallback groq/llama-3.3-70b-versatile failed:", fErr.message);
         }
       }
       const geminiKey = process.env.GEMINI_API_KEY || effectiveKey;
@@ -627,14 +633,16 @@ async function dispatchChatCompletion(req, apiKeyOverride) {
         console.warn(`[LLM Gateway] Provider ${provider} rate limited (${error.message}). Falling back to Google Gemini (gemini-2.5-flash)...`);
         try {
           return await handleGeminiRequest(__spreadProps(__spreadValues({}, req), { model: "gemini-2.5-flash" }), effectiveKey);
-        } catch (e) {
+        } catch (fErr) {
+          console.error("[LLM Gateway] Fallback gemini-2.5-flash failed:", fErr.message);
         }
       }
       if (provider !== "huggingface" && (process.env.HF_TOKEN || process.env.HUGGINGFACE_API_KEY)) {
         console.warn(`[LLM Gateway] Provider ${provider} rate limited (${error.message}). Falling back to Hugging Face (Qwen/Qwen2.5-Coder-32B-Instruct)...`);
         try {
           return await handleHuggingFaceChatRequest(__spreadProps(__spreadValues({}, req), { model: "Qwen/Qwen2.5-Coder-32B-Instruct" }), effectiveKey);
-        } catch (e) {
+        } catch (fErr) {
+          console.error("[LLM Gateway] Fallback HuggingFace Qwen failed:", fErr.message);
         }
       }
       if (provider !== "openai" && process.env.OPENAI_API_KEY) {
@@ -3489,12 +3497,13 @@ var OpenAIProvider = class {
       role: "system",
       content: buildSystemContent(basePrompt, context)
     };
+    const safeMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+      role: m.role || "user",
+      content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+    }));
     const formattedMessages = [
       systemMessage,
-      ...messages.map((m) => ({
-        role: m.role,
-        content: m.content
-      }))
+      ...safeMessages
     ];
     const completion = await this.client.chat.completions.create({
       model: this.llmConfig.model,
@@ -3513,12 +3522,13 @@ var OpenAIProvider = class {
         role: "system",
         content: buildSystemContent(basePrompt, context)
       };
+      const safeStreamMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+        role: m.role || "user",
+        content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+      }));
       const formattedMessages = [
         systemMessage,
-        ...messages.map((m) => ({
-          role: m.role,
-          content: m.content
-        }))
+        ...safeStreamMessages
       ];
       const stream = yield new __await(this.client.chat.completions.create({
         model: this.llmConfig.model,
@@ -4049,9 +4059,9 @@ var GeminiProvider = class {
    *   - Messages must strictly alternate between `user` and `model`.
    */
   buildGeminiContents(messages) {
-    return messages.filter((m) => m.role !== "system").map((m) => ({
+    return (messages || []).filter((m) => Boolean(m && typeof m === "object" && m.role !== "system")).map((m) => ({
       role: m.role === "assistant" ? "model" : "user",
-      parts: [{ text: m.content }]
+      parts: [{ text: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : "" }]
     }));
   }
   // -------------------------------------------------------------------------
@@ -4254,12 +4264,13 @@ var GroqProvider = class {
       role: "system",
       content: buildSystemContent(basePrompt, context)
     };
+    const safeMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+      role: m.role || "user",
+      content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+    }));
     const formattedMessages = [
       systemMessage,
-      ...messages.map((m) => ({
-        role: m.role,
-        content: m.content
-      }))
+      ...safeMessages
     ];
     const completion = await this.client.chat.completions.create({
       model: this.llmConfig.model,
@@ -4278,12 +4289,13 @@ var GroqProvider = class {
         role: "system",
         content: buildSystemContent(basePrompt, context)
       };
+      const safeStreamMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+        role: m.role || "user",
+        content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+      }));
       const formattedMessages = [
         systemMessage,
-        ...messages.map((m) => ({
-          role: m.role,
-          content: m.content
-        }))
+        ...safeStreamMessages
       ];
       const stream = yield new __await(this.client.chat.completions.create({
         model: this.llmConfig.model,
@@ -4412,12 +4424,13 @@ var QwenProvider = class {
       role: "system",
       content: buildSystemContent(basePrompt, context)
     };
+    const safeMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+      role: m.role || "user",
+      content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+    }));
     const formattedMessages = [
       systemMessage,
-      ...messages.map((m) => ({
-        role: m.role,
-        content: m.content
-      }))
+      ...safeMessages
     ];
     const completion = await this.client.chat.completions.create({
       model: this.llmConfig.model,
@@ -4436,12 +4449,13 @@ var QwenProvider = class {
         role: "system",
         content: buildSystemContent(basePrompt, context)
       };
+      const safeStreamMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+        role: m.role || "user",
+        content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+      }));
       const formattedMessages = [
         systemMessage,
-        ...messages.map((m) => ({
-          role: m.role,
-          content: m.content
-        }))
+        ...safeStreamMessages
       ];
       const stream = yield new __await(this.client.chat.completions.create({
         model: this.llmConfig.model,
@@ -4621,14 +4635,18 @@ var UniversalLLMAdapter = class {
     });
   }
   async chat(messages, context) {
-    var _a2, _b, _c;
+    var _a2, _b, _c, _d, _e;
     const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
+    const safeMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+      role: m.role || "user",
+      content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+    }));
     const formattedMessages = [
       { role: "system", content: `${this.systemPrompt}
 
 Context:
 ${context != null ? context : "None"}` },
-      ...messages
+      ...safeMessages
     ];
     let payload;
     if (this.opts.chatPayloadTemplate) {
@@ -4646,8 +4664,15 @@ ${context != null ? context : "None"}` },
         temperature: this.temperature
       };
     }
-    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
-    if (isSelfHost || Boolean(process.env.VERCEL)) {
+    try {
+      const { data: data2 } = await this.http.post(path2, payload);
+      const extractPath2 = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
+      const result2 = (_c = resolvePath(data2, extractPath2)) != null ? _c : extractContent(data2);
+      if (result2 !== void 0) {
+        return String(result2);
+      }
+    } catch (httpErr) {
+      console.warn(`[UniversalLLMAdapter] Direct HTTP POST to ${this.baseUrl}${path2} failed (${httpErr.message}). Attempting in-process gateway dispatch fallback...`);
       const _g2 = globalThis;
       let dispatch = _g2.__retrivoraDispatchChat;
       if (typeof dispatch !== "function") {
@@ -4662,28 +4687,22 @@ ${context != null ? context : "None"}` },
         }
       }
       if (typeof dispatch === "function") {
-        try {
-          const res = await dispatch({
-            model: this.model,
-            messages: formattedMessages,
-            max_tokens: this.maxTokens,
-            temperature: this.temperature
-          }, this.apiKey);
-          const content = extractContent(res == null ? void 0 : res.response);
-          if (content !== void 0) {
-            return content;
-          }
-          throw new Error(`[UniversalLLMAdapter] In-process dispatch returned empty content for model: ${this.model}`);
-        } catch (dispatchErr) {
-          throw dispatchErr;
+        const res = await dispatch({
+          model: this.model,
+          messages: formattedMessages,
+          max_tokens: this.maxTokens,
+          temperature: this.temperature
+        }, this.apiKey);
+        const content = extractContent(res == null ? void 0 : res.response);
+        if (content !== void 0) {
+          return content;
         }
-      } else {
-        throw new Error(`[UniversalLLMAdapter] In-process gateway dispatch not registered. Direct self-referential HTTP calls to ${this.baseUrl} are disabled on Vercel to prevent HTTP 508 Loop Detected.`);
       }
+      throw httpErr;
     }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
-    const result = (_c = resolvePath(data, extractPath)) != null ? _c : extractContent(data);
+    const extractPath = (_d = this.opts.responseExtractPath) != null ? _d : "choices[0].message.content";
+    const result = (_e = resolvePath(data, extractPath)) != null ? _e : extractContent(data);
     if (result === void 0) {
       throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
     }
@@ -4700,12 +4719,16 @@ ${context != null ? context : "None"}` },
       const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
       const url = `${this.baseUrl.replace(/\/$/, "")}${path2}`;
       const extractPath = ((_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content").replace("message.content", "delta.content");
+      const safeMessages = (Array.isArray(messages) ? messages : []).filter((m) => Boolean(m && typeof m === "object")).map((m) => ({
+        role: m.role || "user",
+        content: typeof m.content === "string" ? m.content : m.content != null ? String(m.content) : ""
+      }));
       const formattedMessages = [
         { role: "system", content: `${this.systemPrompt}
 
 Context:
 ${context != null ? context : "None"}` },
-        ...messages
+        ...safeMessages
       ];
       let payload;
       if (this.opts.chatPayloadTemplate) {
@@ -4821,7 +4844,7 @@ ${context != null ? context : "None"}` },
     });
   }
   async embed(text) {
-    var _a2, _b, _c, _d;
+    var _a2, _b, _c, _d, _e;
     const path2 = (_a2 = this.opts.embedPath) != null ? _a2 : "/embeddings";
     let payload;
     if (this.opts.embedPayloadTemplate) {
@@ -4835,8 +4858,15 @@ ${context != null ? context : "None"}` },
         input: text
       };
     }
-    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
-    if (isSelfHost || Boolean(process.env.VERCEL)) {
+    try {
+      const { data: data2 } = await this.http.post(path2, payload);
+      const extractPath2 = (_b = this.opts.embedExtractPath) != null ? _b : "data[0].embedding";
+      const vector2 = resolvePath(data2, extractPath2);
+      if (Array.isArray(vector2)) {
+        return vector2;
+      }
+    } catch (httpErr) {
+      console.warn(`[UniversalLLMAdapter] Direct HTTP embedding POST to ${this.baseUrl}${path2} failed (${httpErr.message}). Attempting in-process gateway dispatch fallback...`);
       const _g2 = globalThis;
       let dispatch = _g2.__retrivoraDispatchEmbedding;
       if (typeof dispatch !== "function") {
@@ -4851,24 +4881,18 @@ ${context != null ? context : "None"}` },
         }
       }
       if (typeof dispatch === "function") {
-        try {
-          const res = await dispatch({
-            model: this.model,
-            input: text
-          }, this.apiKey);
-          if ((_c = (_b = res == null ? void 0 : res.data) == null ? void 0 : _b[0]) == null ? void 0 : _c.embedding) {
-            return res.data[0].embedding;
-          }
-          throw new Error(`[UniversalLLMAdapter] In-process embedding dispatch returned invalid vector for model: ${this.model}`);
-        } catch (dispatchErr) {
-          throw dispatchErr;
+        const res = await dispatch({
+          model: this.model,
+          input: text
+        }, this.apiKey);
+        if ((_d = (_c = res == null ? void 0 : res.data) == null ? void 0 : _c[0]) == null ? void 0 : _d.embedding) {
+          return res.data[0].embedding;
         }
-      } else {
-        throw new Error(`[UniversalLLMAdapter] In-process gateway dispatch not registered. Direct self-referential HTTP calls to ${this.baseUrl} are disabled on Vercel to prevent HTTP 508 Loop Detected.`);
       }
+      throw httpErr;
     }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_d = this.opts.embedExtractPath) != null ? _d : "data[0].embedding";
+    const extractPath = (_e = this.opts.embedExtractPath) != null ? _e : "data[0].embedding";
     const vector = resolvePath(data, extractPath);
     if (!Array.isArray(vector)) {
       throw new Error(`[UniversalLLMAdapter] Expected a number array at '${extractPath}' for embeddings.`);
@@ -6849,8 +6873,8 @@ var IntentClassifier = class {
   }
   // ─── Intent Heuristic Checkers ─────────────────────────────────────────────
   static isInformationLookup(query) {
-    if (/^(what|who|explain|define|where|when|why|how|how does|can|could|does|do|is|are|has|have|will|should|would|meaning of)\s+/i.test(query)) {
-      if (!/\b(compare|versus|vs\.?|trend|monthly|revenue|top selling|best)\b/i.test(query)) {
+    if (/^(what|who|explain|explore|define|where|when|why|how|how does|can|could|does|do|is|are|has|have|will|should|would|meaning of|tell me about|show details of|overview of|summarize)\s+/i.test(query)) {
+      if (!/\b(compare|versus|vs\.?|trend|monthly|revenue|top 10|top selling|best|chart|graph|plot|breakdown|analytics)\b/i.test(query)) {
         return true;
       }
     }
@@ -6887,7 +6911,7 @@ var Rule1SpecificInfoRule = class {
   }
   evaluate(context, intent) {
     const q = (context.userQuery || "").toLowerCase().trim();
-    const isFactQuery = intent === "information_lookup" || /^(what|who|explain|define|where|when|why|meaning of)\s+/i.test(q) || /^(what is the price of|price of|cost of)\b/i.test(q);
+    const isFactQuery = intent === "information_lookup" || /^(what|who|explain|explore|define|where|when|why|meaning of|tell me about|show details of|overview of|summarize)\s+/i.test(q) || /^(what is the price of|price of|cost of)\b/i.test(q);
     const isNonVisual = !/\b(compare|versus|vs\.?|trend|monthly|revenue|best|top 10|chart|table|list|grid)\b/i.test(q);
     if (isFactQuery && isNonVisual) {
       return {
@@ -8174,6 +8198,11 @@ ${schemaProfileText}` : ""}`;
   }
   static chooseAutomaticVisualization(data, profile, query) {
     if (profile.records.length === 0) return null;
+    const q = query.toLowerCase().trim();
+    const isExplicitVisualOrAnalytic = /\b(chart|graph|plot|visualize|trend|monthly|revenue|top\s*\d+|breakdown|analytics|compare|versus|vs\.?|histogram|pie|bar|line|distribution|percentage|share|summary of metrics)\b/i.test(q);
+    if (!isExplicitVisualOrAnalytic) {
+      return null;
+    }
     if (profile.dateFields.length > 0 && profile.numericFields.length > 0) {
       return this.transformToLineChart(profile);
     }
@@ -8289,6 +8318,7 @@ ${schemaProfileText}` : ""}`;
     return Array.from(categories);
   }
   static getProductCategory(item) {
+    if (!item) return null;
     const meta = item.metadata || {};
     const metadataCategory = resolveMetadataValue(meta, "category");
     if (this.isUsableCategory(metadataCategory)) return String(metadataCategory).trim();
@@ -8405,6 +8435,7 @@ ${schemaProfileText}` : ""}`;
       if (!fields.has(normalized)) fields.set(normalized, clean);
     };
     data.forEach((item) => {
+      if (!item) return;
       Object.keys(item.metadata || {}).forEach(addField);
       for (const match of (item.content || "").matchAll(/(?:^|\n)\s*([A-Za-z][A-Za-z0-9_ /-]{1,50})\s*[:\-–]\s*([^\n]+)/g)) {
         addField(match[1]);
@@ -8521,6 +8552,7 @@ ${schemaProfileText}` : ""}`;
     return true;
   }
   static extractStockQuantity(item) {
+    if (!item) return null;
     const meta = item.metadata || {};
     const stockValue = resolveMetadataValue(meta, "stock");
     const numericStock = this.toFiniteNumber(stockValue);
@@ -9781,7 +9813,10 @@ ${context}`;
 Focus on extracting the core intent and entities. Do not answer the question, just rewrite it.
 
 History:
-${history.map((m) => `${m.role}: ${m.content}`).join("\n")}
+${(history || []).map((m) => {
+      var _a2;
+      return `${(m == null ? void 0 : m.role) || "user"}: ${(_a2 = m == null ? void 0 : m.content) != null ? _a2 : ""}`;
+    }).join("\n")}
 
 New Question: ${question}
 
@@ -9805,8 +9840,11 @@ Optimized Search Query:`;
     const ns = formatNamespace(namespace != null ? namespace : this.config.projectId);
     try {
       const { sources } = await this.retrieve(query, { namespace: ns, topK: 5 });
-      if (sources.length === 0) return [];
-      const context = sources.map((s) => s.content).join("\n\n---\n\n");
+      if (!sources || sources.length === 0) return [];
+      const context = sources.map((s) => {
+        var _a2;
+        return (_a2 = s == null ? void 0 : s.content) != null ? _a2 : "";
+      }).join("\n\n---\n\n");
       const prompt = `Based on the following snippets from a document, what are at least 5 short, relevant questions a user might ask? 
 Focus on questions that can be answered by the context.
 Keep each question under 10 words and make them very specific to the content.