@retrivora-ai/rag-engine 2.0.5 → 2.0.6

package/dist/handlers/index.js

@@ -3907,7 +3907,7 @@ var UniversalLLMAdapter = class {
     });
   }
   async chat(messages, context) {
-    var _a2, _b;
+    var _a2, _b, _c, _d, _e, _f;
     const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
     const formattedMessages = [
       { role: "system", content: `${this.systemPrompt}
@@ -3932,8 +3932,26 @@ ${context != null ? context : "None"}` },
         temperature: this.temperature
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchChat === "function") {
+          const res = await _g2.__retrivoraDispatchChat({
+            model: this.model,
+            messages: formattedMessages,
+            max_tokens: this.maxTokens,
+            temperature: this.temperature
+          }, this.apiKey);
+          if ((_e = (_d = (_c = (_b = res == null ? void 0 : res.response) == null ? void 0 : _b.choices) == null ? void 0 : _c[0]) == null ? void 0 : _d.message) == null ? void 0 : _e.content) {
+            return res.response.choices[0].message.content;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
+    const extractPath = (_f = this.opts.responseExtractPath) != null ? _f : "choices[0].message.content";
     const result = resolvePath(data, extractPath);
     if (result === void 0) {
       throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
@@ -3947,7 +3965,7 @@ ${context != null ? context : "None"}` },
    */
   chatStream(messages, context) {
     return __asyncGenerator(this, null, function* () {
-      var _a2, _b, _c;
+      var _a2, _b, _c, _d, _e, _f, _g2;
       const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
       const url = `${this.baseUrl.replace(/\/$/, "")}${path2}`;
       const extractPath = ((_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content").replace("message.content", "delta.content");
@@ -3978,19 +3996,45 @@ ${context != null ? context : "None"}` },
           stream: true
         };
       }
-      const response = yield new __await(fetch(url, {
-        method: "POST",
-        headers: this.resolvedHeaders,
-        body: JSON.stringify(payload)
-      }));
-      if (!response.ok) {
-        const errorText = yield new __await(response.text().catch(() => response.statusText));
-        throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+      const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+      let streamBody = null;
+      if (isSelfHost || Boolean(process.env.VERCEL)) {
+        try {
+          const _g3 = globalThis;
+          if (typeof _g3.__retrivoraDispatchChat === "function") {
+            const res = yield new __await(_g3.__retrivoraDispatchChat({
+              model: this.model,
+              messages: formattedMessages,
+              max_tokens: this.maxTokens,
+              temperature: this.temperature,
+              stream: true
+            }, this.apiKey));
+            if (res == null ? void 0 : res.stream) {
+              streamBody = res.stream;
+            } else if ((_f = (_e = (_d = (_c = res == null ? void 0 : res.response) == null ? void 0 : _c.choices) == null ? void 0 : _d[0]) == null ? void 0 : _e.message) == null ? void 0 : _f.content) {
+              yield res.response.choices[0].message.content;
+              return;
+            }
+          }
+        } catch (e) {
+        }
       }
-      if (!response.body) {
-        throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+      if (!streamBody) {
+        const response = yield new __await(fetch(url, {
+          method: "POST",
+          headers: this.resolvedHeaders,
+          body: JSON.stringify(payload)
+        }));
+        if (!response.ok) {
+          const errorText = yield new __await(response.text().catch(() => response.statusText));
+          throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+        }
+        if (!response.body) {
+          throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+        }
+        streamBody = response.body;
       }
-      const reader = response.body.getReader();
+      const reader = streamBody.getReader();
       const decoder = new TextDecoder("utf-8");
       let buffer = "";
       try {
@@ -3999,7 +4043,7 @@ ${context != null ? context : "None"}` },
           if (done) break;
           buffer += decoder.decode(value, { stream: true });
           const lines = buffer.split("\n");
-          buffer = (_c = lines.pop()) != null ? _c : "";
+          buffer = (_g2 = lines.pop()) != null ? _g2 : "";
           for (const line of lines) {
             const trimmed = line.trim();
             if (!trimmed || trimmed === "data: [DONE]") continue;
@@ -4027,7 +4071,7 @@ ${context != null ? context : "None"}` },
     });
   }
   async embed(text) {
-    var _a2, _b;
+    var _a2, _b, _c, _d;
     const path2 = (_a2 = this.opts.embedPath) != null ? _a2 : "/embeddings";
     let payload;
     if (this.opts.embedPayloadTemplate) {
@@ -4041,8 +4085,24 @@ ${context != null ? context : "None"}` },
         input: text
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchEmbedding === "function") {
+          const res = await _g2.__retrivoraDispatchEmbedding({
+            model: this.model,
+            input: text
+          }, this.apiKey);
+          if ((_c = (_b = res == null ? void 0 : res.data) == null ? void 0 : _b[0]) == null ? void 0 : _c.embedding) {
+            return res.data[0].embedding;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.embedExtractPath) != null ? _b : "data[0].embedding";
+    const extractPath = (_d = this.opts.embedExtractPath) != null ? _d : "data[0].embedding";
     const vector = resolvePath(data, extractPath);
     if (!Array.isArray(vector)) {
       throw new Error(`[UniversalLLMAdapter] Expected a number array at '${extractPath}' for embeddings.`);
@@ -8378,8 +8438,11 @@ ${m.content}`).join("\n\n---\n\n");
           fullSources = fullSources.slice(0, topK);
         }
         const rerankMs = performance.now() - rerankStart;
-        let context = fullSources.length ? fullSources.map((m, i) => `[Source ${i + 1}]
-${m.content}`).join("\n\n---\n\n") : "No relevant context found.";
+        let context = fullSources.length ? fullSources.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;
@@ -8463,7 +8526,10 @@ ${context}`;
         if (isSimulatedThinking) {
           finalRestrictionSuffix += "\n\n(IMPORTANT: You must think step-by-step before answering. Write your thinking process inside a `<think>...</think>` block. Write the `<think>` block first, then follow it with your final response. Keep the thinking process thorough and detailed. Example: `<think>Thinking details here...</think>Final answer text here.`)";
         }
-        const hardenedHistory = [...history];
+        const hardenedHistory = (Array.isArray(history) ? history : []).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 userQuestion = { role: "user", content: question + finalRestrictionSuffix };
         const messages = [...hardenedHistory, userQuestion];
         const systemPrompt = (_v = this.config.llm.systemPrompt) != null ? _v : "";
@@ -8610,13 +8676,13 @@ ${context}`;
           rewrittenQuery,
           systemPrompt,
           userPrompt: question + finalRestrictionSuffix,
-          chunks: sources.map((s) => {
-            var _a3;
+          chunks: (sources || []).filter(Boolean).map((s) => {
+            var _a3, _b2;
             return {
               id: s.id,
               score: s.score,
-              content: s.content,
-              metadata: (_a3 = s.metadata) != null ? _a3 : {},
+              content: (_a3 = s == null ? void 0 : s.content) != null ? _a3 : "",
+              metadata: (_b2 = s == null ? void 0 : s.metadata) != null ? _b2 : {},
               namespace: ns
             };
           }),
@@ -9218,6 +9284,7 @@ ${csv.trim()}`);
 // src/core/DatabaseStorage.ts
 var fs = __toESM(require("fs"));
 var path = __toESM(require("path"));
+var os = __toESM(require("os"));
 var DatabaseStorage = class {
   constructor(config) {
     // Dynamic PG Pool
@@ -9226,7 +9293,8 @@ var DatabaseStorage = class {
     this.mongoClient = null;
     this.mongoDbName = "";
     // Filesystem fallback configuration
-    this.fallbackDir = path.join(process.cwd(), ".retrivora");
+    this.isServerless = Boolean(process.env.VERCEL || process.env.AWS_LAMBDA_FUNCTION_NAME || process.env.LAMBDA_TASK_ROOT);
+    this.fallbackDir = this.isServerless ? path.join(os.tmpdir(), ".retrivora") : path.join(process.cwd(), ".retrivora");
     this.historyFile = path.join(this.fallbackDir, "history.json");
     this.feedbackFile = path.join(this.fallbackDir, "feedback.json");
     var _a2, _b, _c, _d, _e;
@@ -9356,10 +9424,27 @@ var DatabaseStorage = class {
   }
   // Helper for FS fallback writes
   writeLocalFile(filePath, data) {
-    if (!fs.existsSync(this.fallbackDir)) {
-      fs.mkdirSync(this.fallbackDir, { recursive: true });
+    try {
+      const dir = path.dirname(filePath);
+      if (!fs.existsSync(dir)) {
+        fs.mkdirSync(dir, { recursive: true });
+      }
+      fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
+    } catch (err) {
+      if (err.code === "EROFS") {
+        try {
+          const tmpDir = path.join(os.tmpdir(), ".retrivora");
+          if (!fs.existsSync(tmpDir)) {
+            fs.mkdirSync(tmpDir, { recursive: true });
+          }
+          const tmpFilePath = path.join(tmpDir, path.basename(filePath));
+          fs.writeFileSync(tmpFilePath, JSON.stringify(data, null, 2), "utf-8");
+        } catch (e) {
+        }
+      } else {
+        console.warn(`[DatabaseStorage] Failed to write local fallback file: ${err.message}`);
+      }
     }
-    fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
   }
   // ─── Message History Operations ──────────────────────────────────────────
   async saveMessage(sessionId, message) {

package/dist/handlers/index.mjs

@@ -3872,7 +3872,7 @@ var UniversalLLMAdapter = class {
     });
   }
   async chat(messages, context) {
-    var _a2, _b;
+    var _a2, _b, _c, _d, _e, _f;
     const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
     const formattedMessages = [
       { role: "system", content: `${this.systemPrompt}
@@ -3897,8 +3897,26 @@ ${context != null ? context : "None"}` },
         temperature: this.temperature
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchChat === "function") {
+          const res = await _g2.__retrivoraDispatchChat({
+            model: this.model,
+            messages: formattedMessages,
+            max_tokens: this.maxTokens,
+            temperature: this.temperature
+          }, this.apiKey);
+          if ((_e = (_d = (_c = (_b = res == null ? void 0 : res.response) == null ? void 0 : _b.choices) == null ? void 0 : _c[0]) == null ? void 0 : _d.message) == null ? void 0 : _e.content) {
+            return res.response.choices[0].message.content;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
+    const extractPath = (_f = this.opts.responseExtractPath) != null ? _f : "choices[0].message.content";
     const result = resolvePath(data, extractPath);
     if (result === void 0) {
       throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
@@ -3912,7 +3930,7 @@ ${context != null ? context : "None"}` },
    */
   chatStream(messages, context) {
     return __asyncGenerator(this, null, function* () {
-      var _a2, _b, _c;
+      var _a2, _b, _c, _d, _e, _f, _g2;
       const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
       const url = `${this.baseUrl.replace(/\/$/, "")}${path2}`;
       const extractPath = ((_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content").replace("message.content", "delta.content");
@@ -3943,19 +3961,45 @@ ${context != null ? context : "None"}` },
           stream: true
         };
       }
-      const response = yield new __await(fetch(url, {
-        method: "POST",
-        headers: this.resolvedHeaders,
-        body: JSON.stringify(payload)
-      }));
-      if (!response.ok) {
-        const errorText = yield new __await(response.text().catch(() => response.statusText));
-        throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+      const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+      let streamBody = null;
+      if (isSelfHost || Boolean(process.env.VERCEL)) {
+        try {
+          const _g3 = globalThis;
+          if (typeof _g3.__retrivoraDispatchChat === "function") {
+            const res = yield new __await(_g3.__retrivoraDispatchChat({
+              model: this.model,
+              messages: formattedMessages,
+              max_tokens: this.maxTokens,
+              temperature: this.temperature,
+              stream: true
+            }, this.apiKey));
+            if (res == null ? void 0 : res.stream) {
+              streamBody = res.stream;
+            } else if ((_f = (_e = (_d = (_c = res == null ? void 0 : res.response) == null ? void 0 : _c.choices) == null ? void 0 : _d[0]) == null ? void 0 : _e.message) == null ? void 0 : _f.content) {
+              yield res.response.choices[0].message.content;
+              return;
+            }
+          }
+        } catch (e) {
+        }
       }
-      if (!response.body) {
-        throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+      if (!streamBody) {
+        const response = yield new __await(fetch(url, {
+          method: "POST",
+          headers: this.resolvedHeaders,
+          body: JSON.stringify(payload)
+        }));
+        if (!response.ok) {
+          const errorText = yield new __await(response.text().catch(() => response.statusText));
+          throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+        }
+        if (!response.body) {
+          throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+        }
+        streamBody = response.body;
       }
-      const reader = response.body.getReader();
+      const reader = streamBody.getReader();
       const decoder = new TextDecoder("utf-8");
       let buffer = "";
       try {
@@ -3964,7 +4008,7 @@ ${context != null ? context : "None"}` },
           if (done) break;
           buffer += decoder.decode(value, { stream: true });
           const lines = buffer.split("\n");
-          buffer = (_c = lines.pop()) != null ? _c : "";
+          buffer = (_g2 = lines.pop()) != null ? _g2 : "";
           for (const line of lines) {
             const trimmed = line.trim();
             if (!trimmed || trimmed === "data: [DONE]") continue;
@@ -3992,7 +4036,7 @@ ${context != null ? context : "None"}` },
     });
   }
   async embed(text) {
-    var _a2, _b;
+    var _a2, _b, _c, _d;
     const path2 = (_a2 = this.opts.embedPath) != null ? _a2 : "/embeddings";
     let payload;
     if (this.opts.embedPayloadTemplate) {
@@ -4006,8 +4050,24 @@ ${context != null ? context : "None"}` },
         input: text
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchEmbedding === "function") {
+          const res = await _g2.__retrivoraDispatchEmbedding({
+            model: this.model,
+            input: text
+          }, this.apiKey);
+          if ((_c = (_b = res == null ? void 0 : res.data) == null ? void 0 : _b[0]) == null ? void 0 : _c.embedding) {
+            return res.data[0].embedding;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.embedExtractPath) != null ? _b : "data[0].embedding";
+    const extractPath = (_d = this.opts.embedExtractPath) != null ? _d : "data[0].embedding";
     const vector = resolvePath(data, extractPath);
     if (!Array.isArray(vector)) {
       throw new Error(`[UniversalLLMAdapter] Expected a number array at '${extractPath}' for embeddings.`);
@@ -8343,8 +8403,11 @@ ${m.content}`).join("\n\n---\n\n");
           fullSources = fullSources.slice(0, topK);
         }
         const rerankMs = performance.now() - rerankStart;
-        let context = fullSources.length ? fullSources.map((m, i) => `[Source ${i + 1}]
-${m.content}`).join("\n\n---\n\n") : "No relevant context found.";
+        let context = fullSources.length ? fullSources.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;
@@ -8428,7 +8491,10 @@ ${context}`;
         if (isSimulatedThinking) {
           finalRestrictionSuffix += "\n\n(IMPORTANT: You must think step-by-step before answering. Write your thinking process inside a `<think>...</think>` block. Write the `<think>` block first, then follow it with your final response. Keep the thinking process thorough and detailed. Example: `<think>Thinking details here...</think>Final answer text here.`)";
         }
-        const hardenedHistory = [...history];
+        const hardenedHistory = (Array.isArray(history) ? history : []).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 userQuestion = { role: "user", content: question + finalRestrictionSuffix };
         const messages = [...hardenedHistory, userQuestion];
         const systemPrompt = (_v = this.config.llm.systemPrompt) != null ? _v : "";
@@ -8575,13 +8641,13 @@ ${context}`;
           rewrittenQuery,
           systemPrompt,
           userPrompt: question + finalRestrictionSuffix,
-          chunks: sources.map((s) => {
-            var _a3;
+          chunks: (sources || []).filter(Boolean).map((s) => {
+            var _a3, _b2;
             return {
               id: s.id,
               score: s.score,
-              content: s.content,
-              metadata: (_a3 = s.metadata) != null ? _a3 : {},
+              content: (_a3 = s == null ? void 0 : s.content) != null ? _a3 : "",
+              metadata: (_b2 = s == null ? void 0 : s.metadata) != null ? _b2 : {},
               namespace: ns
             };
           }),
@@ -9183,6 +9249,7 @@ ${csv.trim()}`);
 // src/core/DatabaseStorage.ts
 import * as fs from "fs";
 import * as path from "path";
+import * as os from "os";
 var DatabaseStorage = class {
   constructor(config) {
     // Dynamic PG Pool
@@ -9191,7 +9258,8 @@ var DatabaseStorage = class {
     this.mongoClient = null;
     this.mongoDbName = "";
     // Filesystem fallback configuration
-    this.fallbackDir = path.join(process.cwd(), ".retrivora");
+    this.isServerless = Boolean(process.env.VERCEL || process.env.AWS_LAMBDA_FUNCTION_NAME || process.env.LAMBDA_TASK_ROOT);
+    this.fallbackDir = this.isServerless ? path.join(os.tmpdir(), ".retrivora") : path.join(process.cwd(), ".retrivora");
     this.historyFile = path.join(this.fallbackDir, "history.json");
     this.feedbackFile = path.join(this.fallbackDir, "feedback.json");
     var _a2, _b, _c, _d, _e;
@@ -9321,10 +9389,27 @@ var DatabaseStorage = class {
   }
   // Helper for FS fallback writes
   writeLocalFile(filePath, data) {
-    if (!fs.existsSync(this.fallbackDir)) {
-      fs.mkdirSync(this.fallbackDir, { recursive: true });
+    try {
+      const dir = path.dirname(filePath);
+      if (!fs.existsSync(dir)) {
+        fs.mkdirSync(dir, { recursive: true });
+      }
+      fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
+    } catch (err) {
+      if (err.code === "EROFS") {
+        try {
+          const tmpDir = path.join(os.tmpdir(), ".retrivora");
+          if (!fs.existsSync(tmpDir)) {
+            fs.mkdirSync(tmpDir, { recursive: true });
+          }
+          const tmpFilePath = path.join(tmpDir, path.basename(filePath));
+          fs.writeFileSync(tmpFilePath, JSON.stringify(data, null, 2), "utf-8");
+        } catch (e) {
+        }
+      } else {
+        console.warn(`[DatabaseStorage] Failed to write local fallback file: ${err.message}`);
+      }
     }
-    fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
   }
   // ─── Message History Operations ──────────────────────────────────────────
   async saveMessage(sessionId, message) {

package/dist/server.js

@@ -4007,7 +4007,7 @@ var UniversalLLMAdapter = class {
     });
   }
   async chat(messages, context) {
-    var _a2, _b;
+    var _a2, _b, _c, _d, _e, _f;
     const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
     const formattedMessages = [
       { role: "system", content: `${this.systemPrompt}
@@ -4032,8 +4032,26 @@ ${context != null ? context : "None"}` },
         temperature: this.temperature
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchChat === "function") {
+          const res = await _g2.__retrivoraDispatchChat({
+            model: this.model,
+            messages: formattedMessages,
+            max_tokens: this.maxTokens,
+            temperature: this.temperature
+          }, this.apiKey);
+          if ((_e = (_d = (_c = (_b = res == null ? void 0 : res.response) == null ? void 0 : _b.choices) == null ? void 0 : _c[0]) == null ? void 0 : _d.message) == null ? void 0 : _e.content) {
+            return res.response.choices[0].message.content;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
+    const extractPath = (_f = this.opts.responseExtractPath) != null ? _f : "choices[0].message.content";
     const result = resolvePath(data, extractPath);
     if (result === void 0) {
       throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
@@ -4047,7 +4065,7 @@ ${context != null ? context : "None"}` },
    */
   chatStream(messages, context) {
     return __asyncGenerator(this, null, function* () {
-      var _a2, _b, _c;
+      var _a2, _b, _c, _d, _e, _f, _g2;
       const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
       const url = `${this.baseUrl.replace(/\/$/, "")}${path2}`;
       const extractPath = ((_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content").replace("message.content", "delta.content");
@@ -4078,19 +4096,45 @@ ${context != null ? context : "None"}` },
           stream: true
         };
       }
-      const response = yield new __await(fetch(url, {
-        method: "POST",
-        headers: this.resolvedHeaders,
-        body: JSON.stringify(payload)
-      }));
-      if (!response.ok) {
-        const errorText = yield new __await(response.text().catch(() => response.statusText));
-        throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+      const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+      let streamBody = null;
+      if (isSelfHost || Boolean(process.env.VERCEL)) {
+        try {
+          const _g3 = globalThis;
+          if (typeof _g3.__retrivoraDispatchChat === "function") {
+            const res = yield new __await(_g3.__retrivoraDispatchChat({
+              model: this.model,
+              messages: formattedMessages,
+              max_tokens: this.maxTokens,
+              temperature: this.temperature,
+              stream: true
+            }, this.apiKey));
+            if (res == null ? void 0 : res.stream) {
+              streamBody = res.stream;
+            } else if ((_f = (_e = (_d = (_c = res == null ? void 0 : res.response) == null ? void 0 : _c.choices) == null ? void 0 : _d[0]) == null ? void 0 : _e.message) == null ? void 0 : _f.content) {
+              yield res.response.choices[0].message.content;
+              return;
+            }
+          }
+        } catch (e) {
+        }
       }
-      if (!response.body) {
-        throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+      if (!streamBody) {
+        const response = yield new __await(fetch(url, {
+          method: "POST",
+          headers: this.resolvedHeaders,
+          body: JSON.stringify(payload)
+        }));
+        if (!response.ok) {
+          const errorText = yield new __await(response.text().catch(() => response.statusText));
+          throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+        }
+        if (!response.body) {
+          throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+        }
+        streamBody = response.body;
       }
-      const reader = response.body.getReader();
+      const reader = streamBody.getReader();
       const decoder = new TextDecoder("utf-8");
       let buffer = "";
       try {
@@ -4099,7 +4143,7 @@ ${context != null ? context : "None"}` },
           if (done) break;
           buffer += decoder.decode(value, { stream: true });
           const lines = buffer.split("\n");
-          buffer = (_c = lines.pop()) != null ? _c : "";
+          buffer = (_g2 = lines.pop()) != null ? _g2 : "";
           for (const line of lines) {
             const trimmed = line.trim();
             if (!trimmed || trimmed === "data: [DONE]") continue;
@@ -4127,7 +4171,7 @@ ${context != null ? context : "None"}` },
     });
   }
   async embed(text) {
-    var _a2, _b;
+    var _a2, _b, _c, _d;
     const path2 = (_a2 = this.opts.embedPath) != null ? _a2 : "/embeddings";
     let payload;
     if (this.opts.embedPayloadTemplate) {
@@ -4141,8 +4185,24 @@ ${context != null ? context : "None"}` },
         input: text
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchEmbedding === "function") {
+          const res = await _g2.__retrivoraDispatchEmbedding({
+            model: this.model,
+            input: text
+          }, this.apiKey);
+          if ((_c = (_b = res == null ? void 0 : res.data) == null ? void 0 : _b[0]) == null ? void 0 : _c.embedding) {
+            return res.data[0].embedding;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.embedExtractPath) != null ? _b : "data[0].embedding";
+    const extractPath = (_d = this.opts.embedExtractPath) != null ? _d : "data[0].embedding";
     const vector = resolvePath(data, extractPath);
     if (!Array.isArray(vector)) {
       throw new Error(`[UniversalLLMAdapter] Expected a number array at '${extractPath}' for embeddings.`);
@@ -8492,8 +8552,11 @@ ${m.content}`).join("\n\n---\n\n");
           fullSources = fullSources.slice(0, topK);
         }
         const rerankMs = performance.now() - rerankStart;
-        let context = fullSources.length ? fullSources.map((m, i) => `[Source ${i + 1}]
-${m.content}`).join("\n\n---\n\n") : "No relevant context found.";
+        let context = fullSources.length ? fullSources.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;
@@ -8577,7 +8640,10 @@ ${context}`;
         if (isSimulatedThinking) {
           finalRestrictionSuffix += "\n\n(IMPORTANT: You must think step-by-step before answering. Write your thinking process inside a `<think>...</think>` block. Write the `<think>` block first, then follow it with your final response. Keep the thinking process thorough and detailed. Example: `<think>Thinking details here...</think>Final answer text here.`)";
         }
-        const hardenedHistory = [...history];
+        const hardenedHistory = (Array.isArray(history) ? history : []).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 userQuestion = { role: "user", content: question + finalRestrictionSuffix };
         const messages = [...hardenedHistory, userQuestion];
         const systemPrompt = (_v = this.config.llm.systemPrompt) != null ? _v : "";
@@ -8724,13 +8790,13 @@ ${context}`;
           rewrittenQuery,
           systemPrompt,
           userPrompt: question + finalRestrictionSuffix,
-          chunks: sources.map((s) => {
-            var _a3;
+          chunks: (sources || []).filter(Boolean).map((s) => {
+            var _a3, _b2;
             return {
               id: s.id,
               score: s.score,
-              content: s.content,
-              metadata: (_a3 = s.metadata) != null ? _a3 : {},
+              content: (_a3 = s == null ? void 0 : s.content) != null ? _a3 : "",
+              metadata: (_b2 = s == null ? void 0 : s.metadata) != null ? _b2 : {},
               namespace: ns
             };
           }),
@@ -9631,6 +9697,7 @@ var import_server = require("next/server");
 // src/core/DatabaseStorage.ts
 var fs = __toESM(require("fs"));
 var path = __toESM(require("path"));
+var os = __toESM(require("os"));
 var DatabaseStorage = class {
   constructor(config) {
     // Dynamic PG Pool
@@ -9639,7 +9706,8 @@ var DatabaseStorage = class {
     this.mongoClient = null;
     this.mongoDbName = "";
     // Filesystem fallback configuration
-    this.fallbackDir = path.join(process.cwd(), ".retrivora");
+    this.isServerless = Boolean(process.env.VERCEL || process.env.AWS_LAMBDA_FUNCTION_NAME || process.env.LAMBDA_TASK_ROOT);
+    this.fallbackDir = this.isServerless ? path.join(os.tmpdir(), ".retrivora") : path.join(process.cwd(), ".retrivora");
     this.historyFile = path.join(this.fallbackDir, "history.json");
     this.feedbackFile = path.join(this.fallbackDir, "feedback.json");
     var _a2, _b, _c, _d, _e;
@@ -9769,10 +9837,27 @@ var DatabaseStorage = class {
   }
   // Helper for FS fallback writes
   writeLocalFile(filePath, data) {
-    if (!fs.existsSync(this.fallbackDir)) {
-      fs.mkdirSync(this.fallbackDir, { recursive: true });
+    try {
+      const dir = path.dirname(filePath);
+      if (!fs.existsSync(dir)) {
+        fs.mkdirSync(dir, { recursive: true });
+      }
+      fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
+    } catch (err) {
+      if (err.code === "EROFS") {
+        try {
+          const tmpDir = path.join(os.tmpdir(), ".retrivora");
+          if (!fs.existsSync(tmpDir)) {
+            fs.mkdirSync(tmpDir, { recursive: true });
+          }
+          const tmpFilePath = path.join(tmpDir, path.basename(filePath));
+          fs.writeFileSync(tmpFilePath, JSON.stringify(data, null, 2), "utf-8");
+        } catch (e) {
+        }
+      } else {
+        console.warn(`[DatabaseStorage] Failed to write local fallback file: ${err.message}`);
+      }
     }
-    fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
   }
   // ─── Message History Operations ──────────────────────────────────────────
   async saveMessage(sessionId, message) {

package/dist/server.mjs

@@ -3911,7 +3911,7 @@ var UniversalLLMAdapter = class {
     });
   }
   async chat(messages, context) {
-    var _a2, _b;
+    var _a2, _b, _c, _d, _e, _f;
     const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
     const formattedMessages = [
       { role: "system", content: `${this.systemPrompt}
@@ -3936,8 +3936,26 @@ ${context != null ? context : "None"}` },
         temperature: this.temperature
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchChat === "function") {
+          const res = await _g2.__retrivoraDispatchChat({
+            model: this.model,
+            messages: formattedMessages,
+            max_tokens: this.maxTokens,
+            temperature: this.temperature
+          }, this.apiKey);
+          if ((_e = (_d = (_c = (_b = res == null ? void 0 : res.response) == null ? void 0 : _b.choices) == null ? void 0 : _c[0]) == null ? void 0 : _d.message) == null ? void 0 : _e.content) {
+            return res.response.choices[0].message.content;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content";
+    const extractPath = (_f = this.opts.responseExtractPath) != null ? _f : "choices[0].message.content";
     const result = resolvePath(data, extractPath);
     if (result === void 0) {
       throw new Error(`[UniversalLLMAdapter] Could not extract text from path '${extractPath}' in response.`);
@@ -3951,7 +3969,7 @@ ${context != null ? context : "None"}` },
    */
   chatStream(messages, context) {
     return __asyncGenerator(this, null, function* () {
-      var _a2, _b, _c;
+      var _a2, _b, _c, _d, _e, _f, _g2;
       const path2 = (_a2 = this.opts.chatPath) != null ? _a2 : "/chat/completions";
       const url = `${this.baseUrl.replace(/\/$/, "")}${path2}`;
       const extractPath = ((_b = this.opts.responseExtractPath) != null ? _b : "choices[0].message.content").replace("message.content", "delta.content");
@@ -3982,19 +4000,45 @@ ${context != null ? context : "None"}` },
           stream: true
         };
       }
-      const response = yield new __await(fetch(url, {
-        method: "POST",
-        headers: this.resolvedHeaders,
-        body: JSON.stringify(payload)
-      }));
-      if (!response.ok) {
-        const errorText = yield new __await(response.text().catch(() => response.statusText));
-        throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+      const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+      let streamBody = null;
+      if (isSelfHost || Boolean(process.env.VERCEL)) {
+        try {
+          const _g3 = globalThis;
+          if (typeof _g3.__retrivoraDispatchChat === "function") {
+            const res = yield new __await(_g3.__retrivoraDispatchChat({
+              model: this.model,
+              messages: formattedMessages,
+              max_tokens: this.maxTokens,
+              temperature: this.temperature,
+              stream: true
+            }, this.apiKey));
+            if (res == null ? void 0 : res.stream) {
+              streamBody = res.stream;
+            } else if ((_f = (_e = (_d = (_c = res == null ? void 0 : res.response) == null ? void 0 : _c.choices) == null ? void 0 : _d[0]) == null ? void 0 : _e.message) == null ? void 0 : _f.content) {
+              yield res.response.choices[0].message.content;
+              return;
+            }
+          }
+        } catch (e) {
+        }
       }
-      if (!response.body) {
-        throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+      if (!streamBody) {
+        const response = yield new __await(fetch(url, {
+          method: "POST",
+          headers: this.resolvedHeaders,
+          body: JSON.stringify(payload)
+        }));
+        if (!response.ok) {
+          const errorText = yield new __await(response.text().catch(() => response.statusText));
+          throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+        }
+        if (!response.body) {
+          throw new Error("[UniversalLLMAdapter] Response body is null \u2014 server did not send a streaming response.");
+        }
+        streamBody = response.body;
       }
-      const reader = response.body.getReader();
+      const reader = streamBody.getReader();
       const decoder = new TextDecoder("utf-8");
       let buffer = "";
       try {
@@ -4003,7 +4047,7 @@ ${context != null ? context : "None"}` },
           if (done) break;
           buffer += decoder.decode(value, { stream: true });
           const lines = buffer.split("\n");
-          buffer = (_c = lines.pop()) != null ? _c : "";
+          buffer = (_g2 = lines.pop()) != null ? _g2 : "";
           for (const line of lines) {
             const trimmed = line.trim();
             if (!trimmed || trimmed === "data: [DONE]") continue;
@@ -4031,7 +4075,7 @@ ${context != null ? context : "None"}` },
     });
   }
   async embed(text) {
-    var _a2, _b;
+    var _a2, _b, _c, _d;
     const path2 = (_a2 = this.opts.embedPath) != null ? _a2 : "/embeddings";
     let payload;
     if (this.opts.embedPayloadTemplate) {
@@ -4045,8 +4089,24 @@ ${context != null ? context : "None"}` },
         input: text
       };
     }
+    const isSelfHost = this.baseUrl.includes("retrivora.com") || this.baseUrl.includes("localhost");
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g2 = globalThis;
+        if (typeof _g2.__retrivoraDispatchEmbedding === "function") {
+          const res = await _g2.__retrivoraDispatchEmbedding({
+            model: this.model,
+            input: text
+          }, this.apiKey);
+          if ((_c = (_b = res == null ? void 0 : res.data) == null ? void 0 : _b[0]) == null ? void 0 : _c.embedding) {
+            return res.data[0].embedding;
+          }
+        }
+      } catch (e) {
+      }
+    }
     const { data } = await this.http.post(path2, payload);
-    const extractPath = (_b = this.opts.embedExtractPath) != null ? _b : "data[0].embedding";
+    const extractPath = (_d = this.opts.embedExtractPath) != null ? _d : "data[0].embedding";
     const vector = resolvePath(data, extractPath);
     if (!Array.isArray(vector)) {
       throw new Error(`[UniversalLLMAdapter] Expected a number array at '${extractPath}' for embeddings.`);
@@ -8396,8 +8456,11 @@ ${m.content}`).join("\n\n---\n\n");
           fullSources = fullSources.slice(0, topK);
         }
         const rerankMs = performance.now() - rerankStart;
-        let context = fullSources.length ? fullSources.map((m, i) => `[Source ${i + 1}]
-${m.content}`).join("\n\n---\n\n") : "No relevant context found.";
+        let context = fullSources.length ? fullSources.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;
@@ -8481,7 +8544,10 @@ ${context}`;
         if (isSimulatedThinking) {
           finalRestrictionSuffix += "\n\n(IMPORTANT: You must think step-by-step before answering. Write your thinking process inside a `<think>...</think>` block. Write the `<think>` block first, then follow it with your final response. Keep the thinking process thorough and detailed. Example: `<think>Thinking details here...</think>Final answer text here.`)";
         }
-        const hardenedHistory = [...history];
+        const hardenedHistory = (Array.isArray(history) ? history : []).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 userQuestion = { role: "user", content: question + finalRestrictionSuffix };
         const messages = [...hardenedHistory, userQuestion];
         const systemPrompt = (_v = this.config.llm.systemPrompt) != null ? _v : "";
@@ -8628,13 +8694,13 @@ ${context}`;
           rewrittenQuery,
           systemPrompt,
           userPrompt: question + finalRestrictionSuffix,
-          chunks: sources.map((s) => {
-            var _a3;
+          chunks: (sources || []).filter(Boolean).map((s) => {
+            var _a3, _b2;
             return {
               id: s.id,
               score: s.score,
-              content: s.content,
-              metadata: (_a3 = s.metadata) != null ? _a3 : {},
+              content: (_a3 = s == null ? void 0 : s.content) != null ? _a3 : "",
+              metadata: (_b2 = s == null ? void 0 : s.metadata) != null ? _b2 : {},
               namespace: ns
             };
           }),
@@ -9535,6 +9601,7 @@ import { NextResponse } from "next/server";
 // src/core/DatabaseStorage.ts
 import * as fs from "fs";
 import * as path from "path";
+import * as os from "os";
 var DatabaseStorage = class {
   constructor(config) {
     // Dynamic PG Pool
@@ -9543,7 +9610,8 @@ var DatabaseStorage = class {
     this.mongoClient = null;
     this.mongoDbName = "";
     // Filesystem fallback configuration
-    this.fallbackDir = path.join(process.cwd(), ".retrivora");
+    this.isServerless = Boolean(process.env.VERCEL || process.env.AWS_LAMBDA_FUNCTION_NAME || process.env.LAMBDA_TASK_ROOT);
+    this.fallbackDir = this.isServerless ? path.join(os.tmpdir(), ".retrivora") : path.join(process.cwd(), ".retrivora");
     this.historyFile = path.join(this.fallbackDir, "history.json");
     this.feedbackFile = path.join(this.fallbackDir, "feedback.json");
     var _a2, _b, _c, _d, _e;
@@ -9673,10 +9741,27 @@ var DatabaseStorage = class {
   }
   // Helper for FS fallback writes
   writeLocalFile(filePath, data) {
-    if (!fs.existsSync(this.fallbackDir)) {
-      fs.mkdirSync(this.fallbackDir, { recursive: true });
+    try {
+      const dir = path.dirname(filePath);
+      if (!fs.existsSync(dir)) {
+        fs.mkdirSync(dir, { recursive: true });
+      }
+      fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
+    } catch (err) {
+      if (err.code === "EROFS") {
+        try {
+          const tmpDir = path.join(os.tmpdir(), ".retrivora");
+          if (!fs.existsSync(tmpDir)) {
+            fs.mkdirSync(tmpDir, { recursive: true });
+          }
+          const tmpFilePath = path.join(tmpDir, path.basename(filePath));
+          fs.writeFileSync(tmpFilePath, JSON.stringify(data, null, 2), "utf-8");
+        } catch (e) {
+        }
+      } else {
+        console.warn(`[DatabaseStorage] Failed to write local fallback file: ${err.message}`);
+      }
     }
-    fs.writeFileSync(filePath, JSON.stringify(data, null, 2), "utf-8");
   }
   // ─── Message History Operations ──────────────────────────────────────────
   async saveMessage(sessionId, message) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@retrivora-ai/rag-engine",
-  "version": "2.0.5",
+  "version": "2.0.6",
   "description": "Retrivora AI is a plug-and-play AI engine for RAG chat experiences — generic vector DB + LLM provider, embeddable or standalone.",
   "author": "Abhinav Alkuchi",
   "license": "UNLICENSED",

package/src/core/DatabaseStorage.ts

@@ -1,5 +1,6 @@
 import * as fs from 'fs';
 import * as path from 'path';
+import * as os from 'os';
 import { RagConfig } from '../config/RagConfig';
 import { LicenseVerifier } from './LicenseVerifier';
 
@@ -35,7 +36,10 @@ export class DatabaseStorage {
   private feedbackTableName: string;
 
   // Filesystem fallback configuration
-  private fallbackDir = path.join(process.cwd(), '.retrivora');
+  private isServerless = Boolean(process.env.VERCEL || process.env.AWS_LAMBDA_FUNCTION_NAME || process.env.LAMBDA_TASK_ROOT);
+  private fallbackDir = this.isServerless
+    ? path.join(os.tmpdir(), '.retrivora')
+    : path.join(process.cwd(), '.retrivora');
   private historyFile = path.join(this.fallbackDir, 'history.json');
   private feedbackFile = path.join(this.fallbackDir, 'feedback.json');
 
@@ -191,10 +195,28 @@ export class DatabaseStorage {
 
   // Helper for FS fallback writes
   private writeLocalFile(filePath: string, data: any[]): void {
-    if (!fs.existsSync(this.fallbackDir)) {
-      fs.mkdirSync(this.fallbackDir, { recursive: true });
+    try {
+      const dir = path.dirname(filePath);
+      if (!fs.existsSync(dir)) {
+        fs.mkdirSync(dir, { recursive: true });
+      }
+      fs.writeFileSync(filePath, JSON.stringify(data, null, 2), 'utf-8');
+    } catch (err: any) {
+      if (err.code === 'EROFS') {
+        try {
+          const tmpDir = path.join(os.tmpdir(), '.retrivora');
+          if (!fs.existsSync(tmpDir)) {
+            fs.mkdirSync(tmpDir, { recursive: true });
+          }
+          const tmpFilePath = path.join(tmpDir, path.basename(filePath));
+          fs.writeFileSync(tmpFilePath, JSON.stringify(data, null, 2), 'utf-8');
+        } catch {
+          /* ignore write failure in read-only environment */
+        }
+      } else {
+        console.warn(`[DatabaseStorage] Failed to write local fallback file: ${err.message}`);
+      }
     }
-    fs.writeFileSync(filePath, JSON.stringify(data, null, 2), 'utf-8');
   }
 
   // ─── Message History Operations ──────────────────────────────────────────

package/src/core/Pipeline.ts

@@ -608,7 +608,7 @@ You are a helpful assistant. Use the provided context to answer questions accura
 
       // 4. Context Augmentation - Review all retrieved/reranked sources for full comprehension
       let context = fullSources.length
-        ? fullSources.map((m, i) => `[Source ${i + 1}]\n${m.content}`).join('\n\n---\n\n')
+        ? fullSources.map((m, i) => `[Source ${i + 1}]\n${m?.content ?? ''}`).join('\n\n---\n\n')
         : 'No relevant context found.';
 
       // Count the records of relevant sources to pass into the slice dynamically
@@ -722,7 +722,12 @@ You are a helpful assistant. Use the provided context to answer questions accura
         finalRestrictionSuffix += '\n\n(IMPORTANT: You must think step-by-step before answering. Write your thinking process inside a `<think>...</think>` block. Write the `<think>` block first, then follow it with your final response. Keep the thinking process thorough and detailed. Example: `<think>Thinking details here...</think>Final answer text here.`)';
       }
 
-      const hardenedHistory = [...history];
+      const hardenedHistory = (Array.isArray(history) ? history : [])
+        .filter((m): m is ChatMessage => 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 userQuestion = { role: 'user', content: question + finalRestrictionSuffix } as ChatMessage;
       const messages: ChatMessage[] = [...hardenedHistory, userQuestion];
 
@@ -880,11 +885,11 @@ You are a helpful assistant. Use the provided context to answer questions accura
         rewrittenQuery,
         systemPrompt,
         userPrompt: question + finalRestrictionSuffix,
-        chunks: sources.map((s) => ({
+        chunks: (sources || []).filter(Boolean).map((s) => ({
           id: s.id,
           score: s.score,
-          content: s.content,
-          metadata: s.metadata ?? {},
+          content: s?.content ?? '',
+          metadata: s?.metadata ?? {},
           namespace: ns,
         } satisfies RetrievedChunk)),
         latency,

package/src/llm/providers/UniversalLLMAdapter.ts

@@ -94,6 +94,27 @@ export class UniversalLLMAdapter implements ILLMProvider {
       };
     }
 
+    const isSelfHost = this.baseUrl.includes('retrivora.com') || this.baseUrl.includes('localhost');
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g = globalThis as any;
+        if (typeof _g.__retrivoraDispatchChat === 'function') {
+          const res = await _g.__retrivoraDispatchChat({
+            model: this.model,
+            messages: formattedMessages as any,
+            max_tokens: this.maxTokens,
+            temperature: this.temperature,
+          }, this.apiKey);
+
+          if (res?.response?.choices?.[0]?.message?.content) {
+            return res.response.choices[0].message.content;
+          }
+        }
+      } catch {
+        /* proceed to HTTP fetch */
+      }
+    }
+
     const { data } = await this.http.post(path, payload);
 
     const extractPath = this.opts.responseExtractPath ?? 'choices[0].message.content';
@@ -145,22 +166,52 @@ export class UniversalLLMAdapter implements ILLMProvider {
       };
     }
 
-    const response = await fetch(url, {
-      method: 'POST',
-      headers: this.resolvedHeaders,
-      body: JSON.stringify(payload),
-    });
-
-    if (!response.ok) {
-      const errorText = await response.text().catch(() => response.statusText);
-      throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+    const isSelfHost = this.baseUrl.includes('retrivora.com') || this.baseUrl.includes('localhost');
+    let streamBody: ReadableStream<Uint8Array> | null = null;
+
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g = globalThis as any;
+        if (typeof _g.__retrivoraDispatchChat === 'function') {
+          const res = await _g.__retrivoraDispatchChat({
+            model: this.model,
+            messages: formattedMessages as any,
+            max_tokens: this.maxTokens,
+            temperature: this.temperature,
+            stream: true,
+          }, this.apiKey);
+
+          if (res?.stream) {
+            streamBody = res.stream as ReadableStream<Uint8Array>;
+          } else if (res?.response?.choices?.[0]?.message?.content) {
+            yield res.response.choices[0].message.content;
+            return;
+          }
+        }
+      } catch {
+        /* proceed to HTTP fetch */
+      }
     }
 
-    if (!response.body) {
-      throw new Error('[UniversalLLMAdapter] Response body is null — server did not send a streaming response.');
+    if (!streamBody) {
+      const response = await fetch(url, {
+        method: 'POST',
+        headers: this.resolvedHeaders,
+        body: JSON.stringify(payload),
+      });
+
+      if (!response.ok) {
+        const errorText = await response.text().catch(() => response.statusText);
+        throw new Error(`[UniversalLLMAdapter] Streaming request failed (${response.status}): ${errorText}`);
+      }
+
+      if (!response.body) {
+        throw new Error('[UniversalLLMAdapter] Response body is null — server did not send a streaming response.');
+      }
+      streamBody = response.body;
     }
 
-    const reader = response.body.getReader();
+    const reader = streamBody.getReader();
     const decoder = new TextDecoder('utf-8');
     let buffer = '';
 
@@ -218,6 +269,25 @@ export class UniversalLLMAdapter implements ILLMProvider {
       };
     }
 
+    const isSelfHost = this.baseUrl.includes('retrivora.com') || this.baseUrl.includes('localhost');
+    if (isSelfHost || Boolean(process.env.VERCEL)) {
+      try {
+        const _g = globalThis as any;
+        if (typeof _g.__retrivoraDispatchEmbedding === 'function') {
+          const res = await _g.__retrivoraDispatchEmbedding({
+            model: this.model,
+            input: text,
+          }, this.apiKey);
+
+          if (res?.data?.[0]?.embedding) {
+            return res.data[0].embedding;
+          }
+        }
+      } catch {
+        /* proceed to HTTP fetch */
+      }
+    }
+
     const { data } = await this.http.post(path, payload);
 
     const extractPath = this.opts.embedExtractPath ?? 'data[0].embedding';