@retrivora-ai/rag-engine 2.0.7 → 2.0.8

package/dist/server.js

@@ -121,6 +121,609 @@ var init_templateUtils = __esm({
   }
 });
 
+// ../../services/llm-gateway/providers/groq.ts
+async function handleGroqRequest(req, apiKeyOverride) {
+  const apiKey = apiKeyOverride || process.env.GROQ_API_KEY;
+  if (!apiKey) {
+    throw new Error("Groq API Key missing. Please set GROQ_API_KEY in environment variables.");
+  }
+  const modelName = req.model.replace(/^groq\//i, "");
+  const payload = __spreadProps(__spreadValues({}, req), {
+    model: modelName
+  });
+  const response = await fetch("https://api.groq.com/openai/v1/chat/completions", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${apiKey}`
+    },
+    body: JSON.stringify(payload)
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`Groq API Error (${response.status}): ${errorText}`);
+  }
+  if (req.stream && response.body) {
+    return { stream: response.body };
+  }
+  const json = await response.json();
+  return { response: json };
+}
+var init_groq = __esm({
+  "../../services/llm-gateway/providers/groq.ts"() {
+    "use strict";
+  }
+});
+
+// ../../services/llm-gateway/providers/openai.ts
+async function handleOpenAIRequest(req, apiKeyOverride) {
+  const apiKey = apiKeyOverride || process.env.OPENAI_API_KEY;
+  if (!apiKey) {
+    throw new Error("OpenAI API Key missing. Please set OPENAI_API_KEY in environment variables.");
+  }
+  const modelName = req.model.replace(/^(openai|gpt)\//i, "");
+  const payload = __spreadProps(__spreadValues({}, req), {
+    model: modelName
+  });
+  const response = await fetch("https://api.openai.com/v1/chat/completions", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${apiKey}`
+    },
+    body: JSON.stringify(payload)
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`OpenAI API Error (${response.status}): ${errorText}`);
+  }
+  if (req.stream && response.body) {
+    return { stream: response.body };
+  }
+  const json = await response.json();
+  return { response: json };
+}
+async function handleOpenAIEmbedding(req, apiKeyOverride) {
+  const apiKey = apiKeyOverride || process.env.OPENAI_API_KEY;
+  if (!apiKey) {
+    throw new Error("OpenAI API Key missing. Please set OPENAI_API_KEY in environment variables.");
+  }
+  const modelName = req.model.replace(/^(openai)\//i, "");
+  const response = await fetch("https://api.openai.com/v1/embeddings", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${apiKey}`
+    },
+    body: JSON.stringify({
+      model: modelName,
+      input: req.input
+    })
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`OpenAI Embedding API Error (${response.status}): ${errorText}`);
+  }
+  return await response.json();
+}
+var init_openai = __esm({
+  "../../services/llm-gateway/providers/openai.ts"() {
+    "use strict";
+  }
+});
+
+// ../../services/llm-gateway/providers/gemini.ts
+async function handleGeminiRequest(req, apiKeyOverride) {
+  const apiKey = apiKeyOverride || process.env.GEMINI_API_KEY;
+  if (!apiKey) {
+    throw new Error("Gemini API Key missing. Please set GEMINI_API_KEY in environment variables.");
+  }
+  let modelName = req.model.replace(/^(google|gemini)\//i, "");
+  if (!modelName.startsWith("gemini-")) {
+    modelName = `gemini-${modelName}`;
+  }
+  const payload = __spreadProps(__spreadValues({}, req), {
+    model: modelName
+  });
+  const response = await fetch("https://generativelanguage.googleapis.com/v1beta/openai/chat/completions", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${apiKey}`
+    },
+    body: JSON.stringify(payload)
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`Google Gemini API Error (${response.status}): ${errorText}`);
+  }
+  if (req.stream && response.body) {
+    return { stream: response.body };
+  }
+  const json = await response.json();
+  return { response: json };
+}
+async function handleGeminiEmbedding(req, apiKeyOverride) {
+  const apiKey = apiKeyOverride || process.env.GEMINI_API_KEY;
+  if (!apiKey) {
+    throw new Error("Gemini API Key missing. Please set GEMINI_API_KEY in environment variables.");
+  }
+  let modelName = req.model.replace(/^(google|gemini)\//i, "");
+  if (!modelName || modelName === "default" || modelName.includes("text-embedding-3") || modelName.includes("ada") || !modelName.startsWith("text-embedding-00") && !modelName.startsWith("embedding-00")) {
+    modelName = "text-embedding-004";
+  }
+  const response = await fetch("https://generativelanguage.googleapis.com/v1beta/openai/embeddings", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${apiKey}`
+    },
+    body: JSON.stringify({
+      model: modelName,
+      input: req.input
+    })
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`Google Gemini Embedding API Error (${response.status}): ${errorText}`);
+  }
+  return await response.json();
+}
+var init_gemini = __esm({
+  "../../services/llm-gateway/providers/gemini.ts"() {
+    "use strict";
+  }
+});
+
+// ../../services/llm-gateway/providers/huggingface.ts
+async function handleHuggingFaceChatRequest(req, apiKeyOverride) {
+  var _a2, _b;
+  const apiKey = apiKeyOverride || process.env.HUGGINGFACE_API_KEY || process.env.HF_TOKEN;
+  if (!apiKey) {
+    throw new Error("HuggingFace API Key missing. Set HUGGINGFACE_API_KEY or HF_TOKEN in environment variables.");
+  }
+  let modelName = req.model.replace(/^(huggingface|hf)\//i, "");
+  if (!modelName || modelName === "default" || modelName.toLowerCase() === "qwen" || modelName.toLowerCase() === "qwen-coder") {
+    modelName = "Qwen/Qwen2.5-Coder-32B-Instruct";
+  }
+  const payload = __spreadProps(__spreadValues({}, req), {
+    model: modelName
+  });
+  try {
+    const response = await fetch("https://router.huggingface.co/v1/chat/completions", {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        "Authorization": `Bearer ${apiKey}`
+      },
+      body: JSON.stringify(payload)
+    });
+    if (response.ok) {
+      if (req.stream && response.body) {
+        return { stream: response.body };
+      }
+      const json = await response.json();
+      return { response: json };
+    }
+  } catch (e) {
+  }
+  try {
+    const response = await fetch("https://router.huggingface.co/hf-inference/v1/chat/completions", {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        "Authorization": `Bearer ${apiKey}`
+      },
+      body: JSON.stringify(payload)
+    });
+    if (response.ok) {
+      if (req.stream && response.body) {
+        return { stream: response.body };
+      }
+      const json = await response.json();
+      return { response: json };
+    }
+  } catch (e) {
+  }
+  const lastUserMsg = ((_a2 = req.messages.filter((m) => m.role === "user").pop()) == null ? void 0 : _a2.content) || "Hello";
+  const pipelineRes = await fetch(`https://router.huggingface.co/hf-inference/models/${modelName}`, {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "Authorization": `Bearer ${apiKey}`
+    },
+    body: JSON.stringify({
+      inputs: lastUserMsg,
+      parameters: { max_new_tokens: req.max_tokens || 256, temperature: req.temperature || 0.7 },
+      options: { wait_for_model: true }
+    })
+  });
+  if (!pipelineRes.ok) {
+    const errorText = await pipelineRes.text();
+    throw new Error(`HuggingFace API Error (${pipelineRes.status}): ${errorText}`);
+  }
+  const raw = await pipelineRes.json();
+  const textOutput = Array.isArray(raw) ? ((_b = raw[0]) == null ? void 0 : _b.generated_text) || JSON.stringify(raw) : JSON.stringify(raw);
+  const formattedResponse = {
+    id: `hf-${Date.now()}`,
+    object: "chat.completion",
+    created: Math.floor(Date.now() / 1e3),
+    model: modelName,
+    choices: [
+      {
+        index: 0,
+        message: {
+          role: "assistant",
+          content: textOutput
+        },
+        finish_reason: "stop"
+      }
+    ],
+    usage: {
+      prompt_tokens: lastUserMsg.length,
+      completion_tokens: textOutput.length,
+      total_tokens: lastUserMsg.length + textOutput.length
+    }
+  };
+  return { response: formattedResponse };
+}
+async function handleHuggingFaceEmbedding(req, apiKeyOverride) {
+  const apiKey = apiKeyOverride || process.env.HUGGINGFACE_API_KEY || process.env.HF_TOKEN;
+  if (!apiKey) {
+    throw new Error("HuggingFace API Key missing. Set HUGGINGFACE_API_KEY or HF_TOKEN in environment variables.");
+  }
+  let modelName = req.model.replace(/^(huggingface|hf)\//i, "");
+  if (!modelName || modelName === "default" || modelName.includes("text-embedding-3") || modelName.includes("ada") || !modelName.includes("/") && !modelName.includes("bge") && !modelName.includes("sentence-transformers")) {
+    modelName = "BAAI/bge-base-en-v1.5";
+  }
+  const inputs = Array.isArray(req.input) ? req.input : [req.input];
+  const fetchEmbeddingFromModel = async (targetModel) => {
+    try {
+      const routerRes = await fetch("https://router.huggingface.co/hf-inference/v1/embeddings", {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          "Authorization": `Bearer ${apiKey}`
+        },
+        body: JSON.stringify({
+          model: targetModel,
+          input: inputs
+        })
+      });
+      if (routerRes.ok) {
+        const json = await routerRes.json();
+        if (json && json.data) return json;
+      }
+    } catch (e) {
+    }
+    const response = await fetch(`https://router.huggingface.co/hf-inference/models/${targetModel}`, {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        "Authorization": `Bearer ${apiKey}`
+      },
+      body: JSON.stringify({
+        inputs,
+        options: { wait_for_model: true }
+      })
+    });
+    if (!response.ok) {
+      const errorText = await response.text();
+      throw new Error(`HuggingFace API Error (${response.status}): ${errorText}`);
+    }
+    const rawData = await response.json();
+    const embeddings = Array.isArray(rawData[0]) ? typeof rawData[0][0] === "number" ? rawData : rawData.map((arr) => arr[0]) : [rawData];
+    return {
+      object: "list",
+      data: embeddings.map((vec, idx) => ({
+        object: "embedding",
+        embedding: vec,
+        index: idx
+      })),
+      model: targetModel,
+      usage: {
+        prompt_tokens: inputs.reduce((acc, str) => acc + str.length, 0),
+        completion_tokens: 0,
+        total_tokens: inputs.reduce((acc, str) => acc + str.length, 0)
+      }
+    };
+  };
+  try {
+    return await fetchEmbeddingFromModel(modelName);
+  } catch (err) {
+    if (modelName !== "BAAI/bge-base-en-v1.5") {
+      console.warn(`[LLM Gateway] HuggingFace model ${modelName} failed. Falling back to BAAI/bge-base-en-v1.5...`, err);
+      return await fetchEmbeddingFromModel("BAAI/bge-base-en-v1.5");
+    }
+    throw err;
+  }
+}
+var init_huggingface = __esm({
+  "../../services/llm-gateway/providers/huggingface.ts"() {
+    "use strict";
+  }
+});
+
+// ../../services/llm-gateway/providers/anthropic.ts
+async function handleAnthropicRequest(req, apiKeyOverride) {
+  var _a2, _b;
+  const apiKey = apiKeyOverride || process.env.ANTHROPIC_API_KEY;
+  if (!apiKey) {
+    throw new Error("Anthropic API Key missing. Please set ANTHROPIC_API_KEY in environment variables.");
+  }
+  const modelName = req.model.replace(/^(anthropic|claude)\//i, "");
+  let systemPrompt = void 0;
+  const anthropicMessages = [];
+  for (const msg of req.messages) {
+    if (msg.role === "system") {
+      systemPrompt = systemPrompt ? `${systemPrompt}
+${msg.content}` : msg.content;
+    } else {
+      anthropicMessages.push({
+        role: msg.role === "assistant" ? "assistant" : "user",
+        content: msg.content
+      });
+    }
+  }
+  if (anthropicMessages.length === 0) {
+    anthropicMessages.push({ role: "user", content: "Hello" });
+  }
+  const payload = {
+    model: modelName,
+    messages: anthropicMessages,
+    max_tokens: req.max_tokens || req.max_completion_tokens || 1024,
+    stream: req.stream || false
+  };
+  if (systemPrompt) {
+    payload.system = systemPrompt;
+  }
+  if (req.temperature !== void 0) {
+    payload.temperature = req.temperature;
+  }
+  const response = await fetch("https://api.anthropic.com/v1/messages", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      "x-api-key": apiKey,
+      "anthropic-version": "2023-06-01"
+    },
+    body: JSON.stringify(payload)
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`Anthropic API Error (${response.status}): ${errorText}`);
+  }
+  if (req.stream && response.body) {
+    return { stream: response.body };
+  }
+  const json = await response.json();
+  const textContent = ((_b = (_a2 = json.content) == null ? void 0 : _a2[0]) == null ? void 0 : _b.text) || "";
+  const openAIResponse = {
+    id: json.id || `chatcmpl-${Date.now()}`,
+    object: "chat.completion",
+    created: Math.floor(Date.now() / 1e3),
+    model: json.model || modelName,
+    choices: [
+      {
+        index: 0,
+        message: {
+          role: "assistant",
+          content: textContent
+        },
+        finish_reason: json.stop_reason === "end_turn" ? "stop" : json.stop_reason || "stop"
+      }
+    ],
+    usage: json.usage ? {
+      prompt_tokens: json.usage.input_tokens,
+      completion_tokens: json.usage.output_tokens,
+      total_tokens: json.usage.input_tokens + json.usage.output_tokens
+    } : void 0
+  };
+  return { response: openAIResponse };
+}
+var init_anthropic = __esm({
+  "../../services/llm-gateway/providers/anthropic.ts"() {
+    "use strict";
+  }
+});
+
+// ../../services/llm-gateway/providers/ollama.ts
+async function handleOllamaRequest(req, baseUrlOverride) {
+  const baseUrl = (baseUrlOverride || process.env.OLLAMA_BASE_URL || "http://localhost:11434/v1").replace(/\/+$/, "");
+  const modelName = req.model.replace(/^ollama\//i, "");
+  const payload = __spreadProps(__spreadValues({}, req), {
+    model: modelName
+  });
+  const response = await fetch(`${baseUrl}/chat/completions`, {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json"
+    },
+    body: JSON.stringify(payload)
+  });
+  if (!response.ok) {
+    const errorText = await response.text();
+    throw new Error(`Ollama API Error (${response.status}): ${errorText}`);
+  }
+  if (req.stream && response.body) {
+    return { stream: response.body };
+  }
+  const json = await response.json();
+  return { response: json };
+}
+var init_ollama = __esm({
+  "../../services/llm-gateway/providers/ollama.ts"() {
+    "use strict";
+  }
+});
+
+// ../../services/llm-gateway/router.ts
+var router_exports = {};
+__export(router_exports, {
+  SUPPORTED_MODELS: () => SUPPORTED_MODELS,
+  dispatchChatCompletion: () => dispatchChatCompletion,
+  dispatchEmbedding: () => dispatchEmbedding,
+  resolveProvider: () => resolveProvider
+});
+function resolveProvider(model) {
+  const lower = model.toLowerCase();
+  if (lower.startsWith("huggingface/") || lower.startsWith("hf/")) {
+    return "huggingface";
+  }
+  if (lower.startsWith("ollama/")) {
+    return "ollama";
+  }
+  if (lower.startsWith("groq/") || lower.includes("llama") || lower.includes("qwen") || lower.includes("mixtral")) {
+    return "groq";
+  }
+  if (process.env.GROQ_API_KEY) return "groq";
+  if (process.env.OPENAI_API_KEY) return "openai";
+  if (process.env.GEMINI_API_KEY) return "gemini";
+  if (process.env.ANTHROPIC_API_KEY) return "anthropic";
+  if (process.env.HF_TOKEN || process.env.HUGGINGFACE_API_KEY) return "huggingface";
+  return "groq";
+}
+function cleanApiKeyOverride(apiKeyOverride) {
+  if (!apiKeyOverride) return void 0;
+  const key = apiKeyOverride.trim();
+  if (key === process.env.LITELLM_API_KEY || key === process.env.LITELLM_MASTER_KEY || key.startsWith("sk-retrivora") || !key.startsWith("gsk_") && !key.startsWith("AIza") && !key.startsWith("sk-ant-") && !key.startsWith("sk-proj-")) {
+    return void 0;
+  }
+  return key;
+}
+async function dispatchChatCompletion(req, apiKeyOverride) {
+  const effectiveKey = cleanApiKeyOverride(apiKeyOverride);
+  const provider = resolveProvider(req.model);
+  try {
+    switch (provider) {
+      case "groq":
+        return await handleGroqRequest(req, effectiveKey);
+      case "openai":
+        return await handleOpenAIRequest(req, effectiveKey);
+      case "gemini":
+        return await handleGeminiRequest(req, effectiveKey);
+      case "anthropic":
+        return await handleAnthropicRequest(req, effectiveKey);
+      case "ollama":
+        return await handleOllamaRequest(req);
+      case "huggingface":
+        return await handleHuggingFaceChatRequest(req, effectiveKey);
+      default:
+        throw new Error(`Unsupported LLM provider for model: ${req.model}`);
+    }
+  } catch (error) {
+    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();
+      if (isGroqKey && !reqModelLower.includes("llama-3.3-70b")) {
+        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) {
+        }
+      }
+      const geminiKey = process.env.GEMINI_API_KEY || effectiveKey;
+      if (provider !== "gemini" && geminiKey && (geminiKey.startsWith("AIza") || !geminiKey.startsWith("AQ."))) {
+        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) {
+        }
+      }
+      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) {
+        }
+      }
+      if (provider !== "openai" && process.env.OPENAI_API_KEY) {
+        console.warn(`[LLM Gateway] Provider ${provider} rate limited (${error.message}). Falling back to OpenAI (gpt-4o-mini)...`);
+        try {
+          return await handleOpenAIRequest(__spreadProps(__spreadValues({}, req), { model: "gpt-4o-mini" }), effectiveKey);
+        } catch (e) {
+        }
+      }
+      console.warn(`[LLM Gateway] Provider ${provider} rate limited (${error.message}). Falling back to local Ollama (llama3.2)...`);
+      return handleOllamaRequest(__spreadProps(__spreadValues({}, req), { model: "llama3.2" }));
+    }
+    throw error;
+  }
+}
+async function dispatchEmbedding(req, apiKeyOverride) {
+  const effectiveKey = cleanApiKeyOverride(apiKeyOverride);
+  if (req.model.startsWith("hf/") || req.model.startsWith("huggingface/") || req.model.includes("bge-")) {
+    try {
+      return await handleHuggingFaceEmbedding(req, effectiveKey);
+    } catch (err) {
+      console.warn("[LLM Gateway] HuggingFace explicit model embedding failed, falling back to default gateway chain:", err);
+    }
+  }
+  const geminiKey = process.env.GEMINI_API_KEY || effectiveKey;
+  const isGeminiKeyValid = Boolean(geminiKey && (geminiKey.startsWith("AIza") || !geminiKey.startsWith("AQ.")));
+  if (isGeminiKeyValid) {
+    try {
+      return await handleGeminiEmbedding(req, effectiveKey);
+    } catch (err) {
+      console.warn("[LLM Gateway] Gemini embedding failed:", err);
+    }
+  }
+  if (process.env.OPENAI_API_KEY) {
+    try {
+      return await handleOpenAIEmbedding(req, effectiveKey);
+    } catch (err) {
+      console.warn("[LLM Gateway] OpenAI embedding failed:", err);
+    }
+  }
+  if (process.env.HUGGINGFACE_API_KEY || process.env.HF_TOKEN) {
+    try {
+      return await handleHuggingFaceEmbedding(req, effectiveKey);
+    } catch (err) {
+      console.warn("[LLM Gateway] HuggingFace fallback embedding failed:", err);
+    }
+  }
+  return handleGeminiEmbedding(req, effectiveKey);
+}
+var SUPPORTED_MODELS;
+var init_router = __esm({
+  "../../services/llm-gateway/router.ts"() {
+    "use strict";
+    init_groq();
+    init_openai();
+    init_gemini();
+    init_huggingface();
+    init_anthropic();
+    init_ollama();
+    SUPPORTED_MODELS = [
+      { id: "groq/llama-3.1-8b-instant", object: "model", created: 17e8, owned_by: "groq" },
+      { id: "llama-3.1-8b-instant", object: "model", created: 17e8, owned_by: "groq" },
+      { id: "groq/llama-3.3-70b-versatile", object: "model", created: 17e8, owned_by: "groq" },
+      { id: "llama-3.3-70b-versatile", object: "model", created: 17e8, owned_by: "groq" },
+      { id: "gpt-4o-mini", object: "model", created: 17e8, owned_by: "openai" },
+      { id: "openai/gpt-4o-mini", object: "model", created: 17e8, owned_by: "openai" },
+      { id: "gpt-4o", object: "model", created: 17e8, owned_by: "openai" },
+      { id: "gemini-2.5-flash", object: "model", created: 17e8, owned_by: "google" },
+      { id: "google/gemini-2.5-flash", object: "model", created: 17e8, owned_by: "google" },
+      { id: "claude-3-5-haiku-20241022", object: "model", created: 17e8, owned_by: "anthropic" },
+      { id: "anthropic/claude-3-5-haiku-20241022", object: "model", created: 17e8, owned_by: "anthropic" },
+      { id: "text-embedding-004", object: "model", created: 17e8, owned_by: "google" },
+      { id: "google/text-embedding-004", object: "model", created: 17e8, owned_by: "google" },
+      { id: "BAAI/bge-base-en-v1.5", object: "model", created: 17e8, owned_by: "huggingface" },
+      { id: "huggingface/BAAI/bge-base-en-v1.5", object: "model", created: 17e8, owned_by: "huggingface" },
+      { id: "groq/qwen3-32b", object: "model", created: 17e8, owned_by: "groq" },
+      { id: "qwen/qwen3-32b", object: "model", created: 17e8, owned_by: "groq" },
+      { id: "qwen/qwen3-embedding-8b", object: "model", created: 17e8, owned_by: "qwen" },
+      { id: "qwen3-embedding-8b", object: "model", created: 17e8, owned_by: "qwen" },
+      { id: "ollama/llama3.2", object: "model", created: 17e8, owned_by: "ollama" }
+    ];
+    if (typeof globalThis !== "undefined") {
+      const _g2 = globalThis;
+      _g2.__retrivoraDispatchChat = dispatchChatCompletion;
+      _g2.__retrivoraDispatchEmbedding = dispatchEmbedding;
+    }
+  }
+});
+
 // src/providers/vectordb/BaseVectorProvider.ts
 var BaseVectorProvider;
 var init_BaseVectorProvider = __esm({
@@ -3976,6 +4579,17 @@ var VECTOR_PROFILES = {
 };
 
 // src/llm/providers/UniversalLLMAdapter.ts
+function extractContent(obj) {
+  var _a2, _b, _c;
+  if (!obj || typeof obj !== "object") return void 0;
+  const choice = (_a2 = obj.choices) == null ? void 0 : _a2[0];
+  if (!choice) return void 0;
+  if (typeof ((_b = choice.message) == null ? void 0 : _b.content) === "string") return choice.message.content;
+  if (typeof ((_c = choice.delta) == null ? void 0 : _c.content) === "string") return choice.delta.content;
+  if (typeof choice.text === "string") return choice.text;
+  if (typeof choice.content === "string") return choice.content;
+  return void 0;
+}
 var UniversalLLMAdapter = class {
   constructor(config) {
     var _a2, _b, _c, _d, _e, _f, _g2, _h;
@@ -4007,7 +4621,7 @@ var UniversalLLMAdapter = class {
     });
   }
   async chat(messages, context) {
-    var _a2, _b, _c, _d, _e, _f;
+    var _a2, _b, _c;
     const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
     const formattedMessages = [
       { role: "system", content: `${this.systemPrompt}
@@ -4035,25 +4649,41 @@ ${context != null ? context : "None"}` },
     const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
     if (isSelfHost || Boolean(process.env.VERCEL)) {
       const _g2 = globalThis;
-      if (typeof _g2.__retrivoraDispatchChat === "function") {
+      let dispatch = _g2.__retrivoraDispatchChat;
+      if (typeof dispatch !== "function") {
         try {
-          const res = await _g2.__retrivoraDispatchChat({
+          const gateway = await Promise.resolve().then(() => (init_router(), router_exports));
+          if (gateway == null ? void 0 : gateway.dispatchChatCompletion) {
+            _g2.__retrivoraDispatchChat = gateway.dispatchChatCompletion;
+            _g2.__retrivoraDispatchEmbedding = gateway.dispatchEmbedding;
+            dispatch = gateway.dispatchChatCompletion;
+          }
+        } catch (e) {
+        }
+      }
+      if (typeof dispatch === "function") {
+        try {
+          const res = await dispatch({
             model: this.model,
             messages: formattedMessages,
             max_tokens: this.maxTokens,
             temperature: this.temperature
           }, this.apiKey);
-          if ((_e = (_d = (_c = (_b = res == null ? void 0 : res.response) == null ? void 0 : _b.choices) == null ? void 0 : _c[0]) == null ? void 0 : _d.message) == null ? void 0 : _e.content) {
-            return res.response.choices[0].message.content;
+          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;
         }
+      } else {
+        throw new Error(`[UniversalLLMAdapter] In-process gateway dispatch not registered. Direct self-referential HTTP calls to ${this.baseUrl} are disabled on Vercel to prevent HTTP 508 Loop Detected.`);
       }
     }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_f = this.opts.responseExtractPath) != null ? _f : "choices[0].message.content";
-    const result = resolvePath(data, extractPath);
+    const extractPath = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
+    const result = (_c = resolvePath(data, extractPath)) != null ? _c : extractContent(data);
     if (result === void 0) {
       throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
     }
@@ -4066,7 +4696,7 @@ ${context != null ? context : "None"}` },
    */
   chatStream(messages, context) {
     return __asyncGenerator(this, null, function* () {
-      var _a2, _b, _c, _d, _e, _f, _g2;
+      var _a2, _b, _c, _d, _e;
       const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
       const url = `${this.baseUrl.replace(/\/$/, "")}${path2}`;
       const extractPath = ((_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content").replace("message.content", "delta.content");
@@ -4100,10 +4730,22 @@ ${context != null ? context : "None"}` },
       const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
       let streamBody = null;
       if (isSelfHost || Boolean(process.env.VERCEL)) {
-        const _g3 = globalThis;
-        if (typeof _g3.__retrivoraDispatchChat === "function") {
+        const _g2 = globalThis;
+        let dispatch = _g2.__retrivoraDispatchChat;
+        if (typeof dispatch !== "function") {
+          try {
+            const gateway = yield new __await(Promise.resolve().then(() => (init_router(), router_exports)));
+            if (gateway == null ? void 0 : gateway.dispatchChatCompletion) {
+              _g2.__retrivoraDispatchChat = gateway.dispatchChatCompletion;
+              _g2.__retrivoraDispatchEmbedding = gateway.dispatchEmbedding;
+              dispatch = gateway.dispatchChatCompletion;
+            }
+          } catch (e) {
+          }
+        }
+        if (typeof dispatch === "function") {
           try {
-            const res = yield new __await(_g3.__retrivoraDispatchChat({
+            const res = yield new __await(dispatch({
               model: this.model,
               messages: formattedMessages,
               max_tokens: this.maxTokens,
@@ -4112,13 +4754,19 @@ ${context != null ? context : "None"}` },
             }, this.apiKey));
             if (res == null ? void 0 : res.stream) {
               streamBody = res.stream;
-            } else if ((_f = (_e = (_d = (_c = res == null ? void 0 : res.response) == null ? void 0 : _c.choices) == null ? void 0 : _d[0]) == null ? void 0 : _e.message) == null ? void 0 : _f.content) {
-              yield res.response.choices[0].message.content;
-              return;
+            } else {
+              const content = extractContent(res == null ? void 0 : res.response);
+              if (content !== void 0) {
+                yield content;
+                return;
+              }
+              throw new Error(`[UniversalLLMAdapter] In-process dispatch stream returned empty response for model: ${this.model}`);
             }
           } catch (dispatchErr) {
             throw dispatchErr;
           }
+        } 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.`);
         }
       }
       if (!streamBody) {
@@ -4145,14 +4793,14 @@ ${context != null ? context : "None"}` },
           if (done) break;
           buffer += decoder.decode(value, { stream: true });
           const lines = buffer.split("\n");
-          buffer = (_g2 = lines.pop()) != null ? _g2 : "";
+          buffer = (_c = lines.pop()) != null ? _c : "";
           for (const line of lines) {
             const trimmed = line.trim();
             if (!trimmed || trimmed === "data: [DONE]") continue;
             if (!trimmed.startsWith("data:")) continue;
             try {
               const json = JSON.parse(trimmed.slice(5).trim());
-              const text = resolvePath(json, extractPath);
+              const text = (_d = resolvePath(json, extractPath)) != null ? _d : extractContent(json);
               if (text && typeof text === "string") yield text;
             } catch (e) {
             }
@@ -4162,7 +4810,7 @@ ${context != null ? context : "None"}` },
           const jsonStr = buffer.replace(/^data:\s*/, "").trim();
           try {
             const json = JSON.parse(jsonStr);
-            const text = resolvePath(json, extractPath);
+            const text = (_e = resolvePath(json, extractPath)) != null ? _e : extractContent(json);
             if (text && typeof text === "string") yield text;
           } catch (e) {
           }
@@ -4190,18 +4838,33 @@ ${context != null ? context : "None"}` },
     const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
     if (isSelfHost || Boolean(process.env.VERCEL)) {
       const _g2 = globalThis;
-      if (typeof _g2.__retrivoraDispatchEmbedding === "function") {
+      let dispatch = _g2.__retrivoraDispatchEmbedding;
+      if (typeof dispatch !== "function") {
         try {
-          const res = await _g2.__retrivoraDispatchEmbedding({
+          const gateway = await Promise.resolve().then(() => (init_router(), router_exports));
+          if (gateway == null ? void 0 : gateway.dispatchEmbedding) {
+            _g2.__retrivoraDispatchChat = gateway.dispatchChatCompletion;
+            _g2.__retrivoraDispatchEmbedding = gateway.dispatchEmbedding;
+            dispatch = gateway.dispatchEmbedding;
+          }
+        } catch (e) {
+        }
+      }
+      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;
         }
+      } 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.`);
       }
     }
     const { data } = await this.http.post(path2, payload);
@@ -7063,7 +7726,7 @@ ${schemaProfileText}` : ""}`;
   }
   // ─── Transform Helpers ────────────────────────────────────────────────────
   static transformToProductCarousel(data, config, trainedSchema) {
-    const products = data.map((item) => this.extractProductInfo(item, config, trainedSchema)).filter((p) => p !== null).slice(0, 15);
+    const products = data.map((item) => this.extractProductInfo(item, config, trainedSchema)).filter((p) => p !== null);
     return {
       type: "product_carousel",
       title: "Recommended Products",
@@ -7443,7 +8106,8 @@ ${schemaProfileText}` : ""}`;
     return fieldCount.size > 2;
   }
   static profileData(data) {
-    const records = data.map((item) => {
+    const records = (data || []).filter((item) => Boolean(item && typeof item === "object")).map((item) => {
+      var _a2, _b;
       const fields2 = {};
       Object.entries(item.metadata || {}).forEach(([key, value]) => {
         const primitive = this.toPrimitive(value);
@@ -7452,8 +8116,8 @@ ${schemaProfileText}` : ""}`;
       if (!fields2.content && item.content) fields2.content = item.content.substring(0, 500);
       return {
         id: item.id,
-        content: item.content,
-        score: item.score,
+        content: (_a2 = item.content) != null ? _a2 : "",
+        score: (_b = item.score) != null ? _b : 0,
         fields: fields2,
         source: item
       };
@@ -8556,7 +9220,7 @@ ${m.content}`).join("\n\n---\n\n");
    */
   askStreamInternal(_0) {
     return __asyncGenerator(this, arguments, function* (question, history = [], namespace) {
-      var _a2, _b, _c, _d, _e, _f, _g2, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A, _B, _C;
+      var _a2, _b, _c, _d, _e, _f, _g2, _h, _i, _j, _k, _l, _m, _n, _o, _p, _q, _r, _s, _t, _u, _v, _w, _x, _y, _z, _A, _B;
       yield new __await(this.initialize());
       const ns = formatNamespace(namespace != null ? namespace : this.config.projectId);
       const topK = (_b = (_a2 = this.config.rag) == null ? void 0 : _a2.topK) != null ? _b : 5;
@@ -8599,8 +9263,8 @@ ${m.content}`).join("\n\n---\n\n");
         const graphData = (strategyResult === "graph" || strategyResult === "both") && this.graphDB && ((_l = this.config.rag) == null ? void 0 : _l.useGraphRetrieval) ? yield new __await(this.graphDB.query(searchQuery)) : void 0;
         const hasMetadataFilter = filter.metadata && Object.keys(filter.metadata).length > 0;
         const hasNumericPredicates = Array.isArray(filter.__numericPredicates) && filter.__numericPredicates.length > 0;
-        const wantsExhaustiveList = hasNumericPredicates || hasMetadataFilter || /\b(list|all|show|provide|give|get|browse|find|search|display)\b/i.test(question);
-        const retrievalLimit = wantsExhaustiveList ? Math.max(topK * 20, 100) : topK * 2;
+        const wantsExhaustiveList = true;
+        const retrievalLimit = Math.max(topK * 50, 1e3);
         const rawSources = ((strategyResult === "vector" || strategyResult === "both") && queryVector && queryVector.length > 0 ? yield new __await(this.vectorDB.query(queryVector, retrievalLimit, ns, filter)) : []).filter((s) => Boolean(s && typeof s === "object"));
         console.log("[Pipeline] Raw vectorDB.query response:", {
           rawCount: rawSources.length,
@@ -8624,40 +9288,29 @@ ${m.content}`).join("\n\n---\n\n");
           var _a3;
           return ((_a3 = m == null ? void 0 : m.score) != null ? _a3 : 0) >= scoreThreshold;
         });
-        const rerankLimit = wantsExhaustiveList ? retrievalLimit : topK;
+        const rerankLimit = Math.max(retrievalLimit, fullSources.length);
         const useReranking = arch !== "naive" && (arch === "retrieve-and-rerank" || ((_m = this.config.rag) == null ? void 0 : _m.useReranking));
         if (!hasNumericPredicates && useReranking) {
           fullSources = (yield new __await(this.reranker.rerank(fullSources, question, rerankLimit))).filter((s) => Boolean(s && typeof s === "object"));
-        } else if (!wantsExhaustiveList) {
-          fullSources = fullSources.slice(0, topK);
         }
         const rerankMs = performance.now() - rerankStart;
-        let context = fullSources.length ? fullSources.map((m, i) => {
+        const totalCount = fullSources.length;
+        const promptSources = fullSources.slice(0, 15);
+        let context = promptSources.length ? `[SEARCH SUMMARY: Found ${totalCount} matching record(s) out of ${rawSources.length} total resources searched in database. Context includes top ${promptSources.length} relevant records]
+
+` + promptSources.map((m, i) => {
           var _a3;
           return `[Source ${i + 1}]
 ${(_a3 = m == null ? void 0 : m.content) != null ? _a3 : ""}`;
         }).join("\n\n---\n\n") : "No relevant context found.";
-        let displayCount = 15;
-        if (hasMetadataFilter) {
-          displayCount = fullSources.length;
-        } else {
-          const baseThreshold = Math.max(scoreThreshold, 0.4);
-          const highlyRelevant = fullSources.filter((m) => {
-            var _a3;
-            return ((_a3 = m == null ? void 0 : m.score) != null ? _a3 : 0) >= baseThreshold;
-          });
-          displayCount = Math.max(highlyRelevant.length, topK);
-          if (displayCount > 15) {
-            displayCount = 15;
-          }
-        }
         const sources = [...fullSources].filter((s) => Boolean(s && typeof s === "object")).sort((a, b) => {
           var _a3, _b2;
           return ((_a3 = b == null ? void 0 : b.score) != null ? _a3 : 0) - ((_b2 = a == null ? void 0 : a.score) != null ? _b2 : 0);
-        }).slice(0, displayCount);
+        });
         console.log("[Pipeline] Final sources for prompt & UI:", {
           count: sources.length,
-          sources: sources.map((s, idx) => {
+          totalRawCount: rawSources.length,
+          sources: sources.slice(0, 10).map((s, idx) => {
             var _a3;
             return {
               rank: idx + 1,
@@ -8857,7 +9510,7 @@ ${context}`;
           }
           yield fullReply;
         }
-        const runHallucination = ((_w = this.config.llm.options) == null ? void 0 : _w.hallucinationScoring) === true || this.config.llm.provider !== "ollama" && ((_x = this.config.llm.options) == null ? void 0 : _x.hallucinationScoring) !== false;
+        const runHallucination = ((_w = this.config.llm.options) == null ? void 0 : _w.hallucinationScoring) === true;
         if (runHallucination) {
           hallucinationScoringPromise = scoreHallucination(this.llmProvider, fullReply, context).catch(() => void 0);
         }
@@ -8866,7 +9519,7 @@ ${context}`;
         const latency = {
           embedMs: Math.round(embedMs),
           retrieveMs: Math.round(retrieveMs),
-          rerankMs: arch !== "naive" && (arch === "retrieve-and-rerank" || ((_y = this.config.rag) == null ? void 0 : _y.useReranking)) ? Math.round(rerankMs) : void 0,
+          rerankMs: arch !== "naive" && (arch === "retrieve-and-rerank" || ((_x = this.config.rag) == null ? void 0 : _x.useReranking)) ? Math.round(rerankMs) : void 0,
           generateMs: Math.round(generateMs),
           totalMs: Math.round(totalMs)
         };
@@ -8879,7 +9532,7 @@ ${context}`;
           totalTokens: promptTokens + completionTokens,
           estimatedCostUsd: estimateCostUsd(promptTokens, completionTokens, this.config.llm.model)
         };
-        const awaitHallucination = ((_z = this.config.llm.options) == null ? void 0 : _z.awaitHallucination) === true;
+        const awaitHallucination = ((_y = this.config.llm.options) == null ? void 0 : _y.awaitHallucination) === true;
         const [uiTransformation, hallucinationResult] = yield new __await(Promise.all([
           uiTransformationPromise,
           awaitHallucination ? hallucinationScoringPromise : Promise.resolve(void 0)
@@ -8908,10 +9561,10 @@ ${context}`;
           hallucinationReason: hScore.reason
         } : {});
         const trace = buildTrace(hallucinationResult);
-        const isTelemetryActive = (_B = (_A = this.config.telemetry) == null ? void 0 : _A.enabled) != null ? _B : Boolean(this.config.licenseKey);
+        const isTelemetryActive = (_A = (_z = this.config.telemetry) == null ? void 0 : _z.enabled) != null ? _A : Boolean(this.config.licenseKey);
         if (isTelemetryActive) {
           const defaultUrl = process.env.NODE_ENV === "development" && !process.env.TELEMETRY_URL && !process.env.NEXT_PUBLIC_TELEMETRY_URL ? "http://localhost:3001/api/telemetry" : "https://retrivora.com/api/telemetry";
-          const telemetryUrl = ((_C = this.config.telemetry) == null ? void 0 : _C.url) || process.env.TELEMETRY_URL || process.env.NEXT_PUBLIC_TELEMETRY_URL || defaultUrl;
+          const telemetryUrl = ((_B = this.config.telemetry) == null ? void 0 : _B.url) || process.env.TELEMETRY_URL || process.env.NEXT_PUBLIC_TELEMETRY_URL || defaultUrl;
           const absoluteUrl = telemetryUrl.startsWith("http") ? telemetryUrl : (process.env.NEXT_PUBLIC_APP_URL || `http://localhost:${process.env.PORT || 3e3}`) + telemetryUrl;
           (async () => {
             try {
@@ -8970,9 +9623,8 @@ ${context}`;
         { visualizationHint: "product_browse", recommendedChart: "text", filterInStockOnly: false, wantsExplicitTable: false, isTemporal: false, isComparison: false, language: "en" }
       );
     }
-    const isLocalProvider = this.config.llm.provider === "ollama";
-    const disableLlmUiTransform = ((_a2 = this.config.llm.options) == null ? void 0 : _a2.disableLlmUiTransform) === true;
-    if (forceDeterministic || isLocalProvider || disableLlmUiTransform) {
+    const enableLlmUiTransform = ((_a2 = this.config.llm.options) == null ? void 0 : _a2.enableLlmUiTransform) === true;
+    if (forceDeterministic || !enableLlmUiTransform) {
       return UITransformer.transform(question, sources, this.config, cachedSchema);
     }
     try {