@retrivora-ai/rag-engine 1.9.8 → 1.9.9

package/dist/server.mjs

@@ -19,6 +19,12 @@ var __spreadValues = (a, b) => {
   return a;
 };
 var __spreadProps = (a, b) => __defProps(a, __getOwnPropDescs(b));
+var __require = /* @__PURE__ */ ((x) => typeof require !== "undefined" ? require : typeof Proxy !== "undefined" ? new Proxy(x, {
+  get: (a, b) => (typeof require !== "undefined" ? require : a)[b]
+}) : x)(function(x) {
+  if (typeof require !== "undefined") return require.apply(this, arguments);
+  throw Error('Dynamic require of "' + x + '" is not supported');
+});
 var __objRest = (source, exclude) => {
   var target = {};
   for (var prop in source)
@@ -2344,7 +2350,7 @@ var ConfigResolver = class {
     }
   }
   static mergeRetrievalWorkflow(rag, retrieval, workflow) {
-    var _a2, _b, _c, _d, _e;
+    var _a2, _b, _c, _d, _e, _f;
     if (!rag && !retrieval && !workflow) return void 0;
     const normalized = __spreadValues({}, rag != null ? rag : {});
     if (retrieval) {
@@ -2352,20 +2358,44 @@ var ConfigResolver = class {
       normalized.scoreThreshold = (_b = retrieval.scoreThreshold) != null ? _b : normalized.scoreThreshold;
       normalized.useReranking = (_c = retrieval.useReranking) != null ? _c : normalized.useReranking;
       if (retrieval.strategy === "hybrid") normalized.architecture = (_d = normalized.architecture) != null ? _d : "hybrid";
-      if (retrieval.strategy === "agentic") normalized.architecture = "agentic";
-      if (retrieval.strategy === "contextual-compression") normalized.useReranking = true;
+      if (retrieval.strategy === "agentic") normalized.architecture = "multi-agent";
+      if (retrieval.strategy === "contextual-compression") {
+        normalized.useReranking = true;
+        normalized.architecture = (_e = normalized.architecture) != null ? _e : "retrieve-and-rerank";
+      }
     }
     if (workflow == null ? void 0 : workflow.type) {
       const type = workflow.type;
-      if (type === "agentic" || type === "agentic-rag") normalized.architecture = "agentic";
-      else if (type === "hybrid-rag") normalized.architecture = "hybrid";
-      else if (type === "graph-rag") {
+      if (type === "naive-rag") {
+        normalized.architecture = "naive";
+        normalized.useReranking = false;
+        normalized.useGraphRetrieval = false;
+      } else if (type === "retrieve-and-rerank") {
+        normalized.architecture = "retrieve-and-rerank";
+        normalized.useReranking = true;
+      } else if (type === "multimodal-rag") {
+        normalized.architecture = "multimodal";
+      } else if (type === "graph-rag") {
         normalized.architecture = "graph";
         normalized.useGraphRetrieval = true;
+      } else if (type === "hybrid-rag") {
+        normalized.architecture = "hybrid";
+      } else if (type === "agentic-router") {
+        normalized.architecture = "agentic-router";
+      } else if (type === "multi-agent-rag" || type === "multi-agent" || type === "agentic" || type === "agentic-rag") {
+        normalized.architecture = "multi-agent";
       } else if (type === "simple-rag" || type === "rag") {
-        normalized.architecture = (_e = normalized.architecture) != null ? _e : "simple";
+        normalized.architecture = (_f = normalized.architecture) != null ? _f : "naive";
       }
     }
+    if (normalized.architecture === "retrieve-and-rerank") {
+      normalized.useReranking = true;
+    } else if (normalized.architecture === "graph") {
+      normalized.useGraphRetrieval = true;
+    } else if (normalized.architecture === "naive") {
+      normalized.useReranking = false;
+      normalized.useGraphRetrieval = false;
+    }
     return normalized;
   }
 };
@@ -4220,6 +4250,19 @@ var LicenseVerifier = class {
    */
   static verify(licenseKey, currentProjectId, provider, publicKeyOverride) {
     const isProduction = process.env.NODE_ENV === "production";
+    const now = Date.now();
+    if (licenseKey) {
+      const cacheKey = `${licenseKey}:${currentProjectId}:${provider || ""}:${publicKeyOverride || ""}`;
+      const cached = this.cache.get(cacheKey);
+      if (cached && now - cached.cachedAt < this.CACHE_TTL_MS) {
+        const currentTimestampSec = Math.floor(now / 1e3);
+        if (cached.payload.expiresAt && currentTimestampSec > cached.payload.expiresAt) {
+          this.cache.delete(cacheKey);
+        } else {
+          return cached.payload;
+        }
+      }
+    }
     if (!licenseKey) {
       if (isProduction) {
         throw new ConfigurationException(
@@ -4306,6 +4349,10 @@ var LicenseVerifier = class {
           }
         }
       }
+      if (licenseKey) {
+        const cacheKey = `${licenseKey}:${currentProjectId}:${provider || ""}:${publicKeyOverride || ""}`;
+        this.cache.set(cacheKey, { payload, cachedAt: now });
+      }
       return payload;
     } catch (err) {
       const message = err instanceof Error ? err.message : String(err);
@@ -4314,7 +4361,76 @@ var LicenseVerifier = class {
       );
     }
   }
+  static async verifyIP(licenseKey) {
+    if (!licenseKey) return;
+    try {
+      const parts = licenseKey.split(".");
+      if (parts.length !== 3) return;
+      const [, payloadB64] = parts;
+      const payload = JSON.parse(
+        Buffer.from(payloadB64, "base64url").toString("utf8")
+      );
+      const tier = (payload.tier || "").toLowerCase();
+      if (tier === "enterprise" || tier === "custom") {
+        return;
+      }
+      if (payload.ipv4 || payload.ipv6) {
+        let currentIpv4 = "";
+        let currentIpv6 = "";
+        try {
+          const res = await fetch("https://api4.ipify.org?format=json", { signal: AbortSignal.timeout(3e3) });
+          const data = await res.json();
+          currentIpv4 = data.ip;
+        } catch (e) {
+        }
+        try {
+          const res = await fetch("https://api6.ipify.org?format=json", { signal: AbortSignal.timeout(3e3) });
+          const data = await res.json();
+          currentIpv6 = data.ip;
+        } catch (e) {
+        }
+        const localIps = [];
+        try {
+          const osModule = __require("os");
+          const interfaces = osModule.networkInterfaces();
+          for (const name of Object.keys(interfaces)) {
+            for (const iface of interfaces[name] || []) {
+              if (!iface.internal) {
+                localIps.push(iface.address);
+              }
+            }
+          }
+        } catch (e) {
+        }
+        const isIpv4Match = payload.ipv4 && (currentIpv4 === payload.ipv4 || localIps.includes(payload.ipv4));
+        const isIpv6Match = payload.ipv6 && (currentIpv6 === payload.ipv6 || localIps.includes(payload.ipv6));
+        if (payload.ipv4 && payload.ipv6) {
+          if (!isIpv4Match && !isIpv6Match) {
+            throw new Error(
+              `License key access restricted. This license key was created for a different system (authorized IPv4: ${payload.ipv4}, IPv6: ${payload.ipv6}) and cannot be used on this machine.`
+            );
+          }
+        } else if (payload.ipv4 && !isIpv4Match) {
+          throw new Error(
+            `License key access restricted. This license key was created for a different system (authorized IPv4: ${payload.ipv4}) and cannot be used on this machine.`
+          );
+        } else if (payload.ipv6 && !isIpv6Match) {
+          throw new Error(
+            `License key access restricted. This license key was created for a different system (authorized IPv6: ${payload.ipv6}) and cannot be used on this machine.`
+          );
+        }
+      }
+    } catch (err) {
+      if (err.message.includes("License key access restricted")) {
+        throw new ConfigurationException(`[Retrivora SDK] ${err.message}`);
+      }
+    }
+  }
 };
+// A simple in-memory cache to save CPU overhead of cryptographic signature checks.
+LicenseVerifier.cache = /* @__PURE__ */ new Map();
+LicenseVerifier.CACHE_TTL_MS = 60 * 1e3;
+// Cache verified license for 60 seconds
 // Retrivora's Public Key used to verify the license key signature.
 // In production, this matches the private key held by Retrivora SaaS.
 LicenseVerifier.PUBLIC_KEY = `-----BEGIN PUBLIC KEY-----
@@ -7275,7 +7391,7 @@ var Pipeline = class {
     return this.initialised ? this.llmProvider : void 0;
   }
   async initialize() {
-    var _a2, _b, _c;
+    var _a2, _b, _c, _d;
     if (this.initialised) return;
     const chartInstruction = `You are a helpful assistant. Use the provided context to answer questions accurately.
 
@@ -7318,7 +7434,7 @@ var Pipeline = class {
       this.entityExtractor = new EntityExtractor(this.llmProvider);
     }
     await this.vectorDB.initialize();
-    if (((_a2 = this.config.rag) == null ? void 0 : _a2.architecture) === "agentic" || this.config.mcpServers && this.config.mcpServers.length > 0) {
+    if (((_a2 = this.config.rag) == null ? void 0 : _a2.architecture) === "agentic" || ((_b = this.config.rag) == null ? void 0 : _b.architecture) === "multi-agent" || this.config.mcpServers && this.config.mcpServers.length > 0) {
       this.mcpRegistry = new MCPRegistry(this.config.mcpServers || []);
       this.multiAgentCoordinator = new MultiAgentCoordinator({
         llmProvider: this.llmProvider,
@@ -7330,7 +7446,7 @@ var Pipeline = class {
       this.agent = new LangChainAgent(this, this.config);
       await this.agent.initialize(this.llmProvider);
     }
-    if ((_c = (_b = this.config.rag) == null ? void 0 : _b.cag) == null ? void 0 : _c.enabled) {
+    if ((_d = (_c = this.config.rag) == null ? void 0 : _c.cag) == null ? void 0 : _d.enabled) {
       await this.loadColdContext(this.config.projectId);
     }
     this.initialised = true;
@@ -7493,9 +7609,9 @@ ${m.content}`).join("\n\n---\n\n");
     return { reply, sources };
   }
   async ask(question, history = [], namespace) {
-    var _a2;
+    var _a2, _b;
     await this.initialize();
-    if ((((_a2 = this.config.rag) == null ? void 0 : _a2.architecture) === "agentic" || this.config.mcpServers && this.config.mcpServers.length > 0) && this.multiAgentCoordinator) {
+    if ((((_a2 = this.config.rag) == null ? void 0 : _a2.architecture) === "agentic" || ((_b = this.config.rag) == null ? void 0 : _b.architecture) === "multi-agent" || this.config.mcpServers && this.config.mcpServers.length > 0) && this.multiAgentCoordinator) {
       return await this.multiAgentCoordinator.run(question, history);
     }
     const stream = this.askStream(question, history, namespace);
@@ -7530,9 +7646,9 @@ ${m.content}`).join("\n\n---\n\n");
   }
   askStream(_0) {
     return __asyncGenerator(this, arguments, function* (question, history = [], namespace) {
-      var _a2;
+      var _a2, _b;
       yield new __await(this.initialize());
-      if ((((_a2 = this.config.rag) == null ? void 0 : _a2.architecture) === "agentic" || this.config.mcpServers && this.config.mcpServers.length > 0) && this.multiAgentCoordinator) {
+      if ((((_a2 = this.config.rag) == null ? void 0 : _a2.architecture) === "agentic" || ((_b = this.config.rag) == null ? void 0 : _b.architecture) === "multi-agent" || this.config.mcpServers && this.config.mcpServers.length > 0) && this.multiAgentCoordinator) {
         const stream2 = this.multiAgentCoordinator.runStream(question, history);
         try {
           for (var iter = __forAwait(stream2), more, temp, error; more = !(temp = yield new __await(iter.next())).done; more = false) {
@@ -7581,7 +7697,7 @@ ${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;
+      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;
       yield new __await(this.initialize());
       const ns = namespace != null ? namespace : this.config.projectId;
       const topK = (_b = (_a2 = this.config.rag) == null ? void 0 : _a2.topK) != null ? _b : 5;
@@ -7604,13 +7720,14 @@ ${m.content}`).join("\n\n---\n\n");
         const embedStart = performance.now();
         const cacheKey = `${ns}::${searchQuery}`;
         const cachedVector = this.embeddingCache.get(cacheKey);
-        const useGraph = this.graphDB && ((_h = this.config.rag) == null ? void 0 : _h.useGraphRetrieval);
+        const arch = (_h = this.config.rag) == null ? void 0 : _h.architecture;
+        const useGraph = arch !== "naive" && arch !== "retrieve-and-rerank" && this.graphDB && ((_i = this.config.rag) == null ? void 0 : _i.useGraphRetrieval);
         const [strategyResult, embeddedVector] = yield new __await(Promise.all([
           QueryProcessor.determineRetrievalStrategy(
             searchQuery,
             useGraph ? this.llmRouter.get("fast") : void 0,
-            (_i = this.config.rag) == null ? void 0 : _i.graphKeywords,
-            (_j = this.config.rag) == null ? void 0 : _j.vectorKeywords
+            (_j = this.config.rag) == null ? void 0 : _j.graphKeywords,
+            (_k = this.config.rag) == null ? void 0 : _k.vectorKeywords
           ),
           // Embed immediately regardless of strategy — costs nothing if cached.
           // If strategy turns out to be 'graph'-only we just won't use the vector.
@@ -7620,7 +7737,7 @@ ${m.content}`).join("\n\n---\n\n");
         if (!cachedVector && queryVector.length > 0) {
           this.embeddingCache.set(cacheKey, queryVector);
         }
-        const graphData = (strategyResult === "graph" || strategyResult === "both") && this.graphDB && ((_k = this.config.rag) == null ? void 0 : _k.useGraphRetrieval) ? yield new __await(this.graphDB.query(searchQuery)) : void 0;
+        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);
@@ -7633,7 +7750,8 @@ ${m.content}`).join("\n\n---\n\n");
         const structuredSources = this.applyStructuredFilters(rawSources, filter);
         let fullSources = hasNumericPredicates ? structuredSources : structuredSources.filter((m) => m.score >= scoreThreshold);
         const rerankLimit = wantsExhaustiveList ? retrievalLimit : topK;
-        if (!hasNumericPredicates && ((_l = this.config.rag) == null ? void 0 : _l.useReranking)) {
+        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));
         } else if (!wantsExhaustiveList) {
           fullSources = fullSources.slice(0, topK);
@@ -7663,7 +7781,7 @@ ${graphContext}
 VECTOR CONTEXT:
 ${context}`;
         }
-        if ((_n = (_m = this.config.rag) == null ? void 0 : _m.cag) == null ? void 0 : _n.enabled) {
+        if ((_o = (_n = this.config.rag) == null ? void 0 : _n.cag) == null ? void 0 : _o.enabled) {
           if (!this.coldContexts.has(ns)) {
             yield new __await(this.loadColdContext(ns));
           }
@@ -7716,10 +7834,10 @@ ${context}`;
         let hallucinationScoringPromise = Promise.resolve(void 0);
         const restrictionSuffix = "\n\n(IMPORTANT: Format your response beautifully using rich Markdown! Use bolding for emphasis, headings (##) for structure, bullet points for lists, and proper line breaks between paragraphs to make it highly readable. However, NEVER generate Markdown tables, HTML figures, or text charts (the UI renders requested charts separately, so NEVER say you cannot create, display, draw, render, or provide a chart/visualization). If listing products, use simple markdown bullet points or comma-separated names. NEVER use plus signs (+) to separate product names, category names, or list items. For product description/detail questions, provide customer-facing prose only and do NOT list internal catalog fields such as Handle, Body (HTML), Vendor, Type, Tags, Published, Option, Variant, SKU, or raw metadata labels. You CAN use numbers for years/counts like 2006 or 5800, but NEVER put a dollar sign ($) before them.)";
         const isClaude37 = this.config.llm.model.includes("claude-3-7-sonnet");
-        const isNativeThinking = isClaude37 && (((_o = this.config.llm.options) == null ? void 0 : _o.thinking) === true || ((_p = this.config.llm.options) == null ? void 0 : _p.thinking) === "enabled" || ((_q = this.config.llm.options) == null ? void 0 : _q.thinking) !== false);
+        const isNativeThinking = isClaude37 && (((_p = this.config.llm.options) == null ? void 0 : _p.thinking) === true || ((_q = this.config.llm.options) == null ? void 0 : _q.thinking) === "enabled" || ((_r = this.config.llm.options) == null ? void 0 : _r.thinking) !== false);
         const modelNameLower = this.config.llm.model.toLowerCase();
         const isReasoningModel = modelNameLower.includes("-r1") || modelNameLower.includes("deepseek-r1") || modelNameLower.includes("reasoning") || modelNameLower.includes("thinking");
-        const isSimulatedThinking = !isNativeThinking && (((_r = this.config.llm.options) == null ? void 0 : _r.thinking) === true || ((_s = this.config.llm.options) == null ? void 0 : _s.thinking) === "enabled" || isReasoningModel && ((_t = this.config.llm.options) == null ? void 0 : _t.thinking) !== false);
+        const isSimulatedThinking = !isNativeThinking && (((_s = this.config.llm.options) == null ? void 0 : _s.thinking) === true || ((_t = this.config.llm.options) == null ? void 0 : _t.thinking) === "enabled" || isReasoningModel && ((_u = this.config.llm.options) == null ? void 0 : _u.thinking) !== false);
         let finalRestrictionSuffix = restrictionSuffix;
         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.`)";
@@ -7727,7 +7845,7 @@ ${context}`;
         const hardenedHistory = [...history];
         const userQuestion = { role: "user", content: question + finalRestrictionSuffix };
         const messages = [...hardenedHistory, userQuestion];
-        const systemPrompt = (_u = this.config.llm.systemPrompt) != null ? _u : "";
+        const systemPrompt = (_v = this.config.llm.systemPrompt) != null ? _v : "";
         const userPrompt = messages.map((m) => `${m.role}: ${m.content}`).join("\n");
         let fullReply = "";
         let thinkingText = "";
@@ -7838,7 +7956,7 @@ ${context}`;
           }
           yield fullReply;
         }
-        const runHallucination = ((_v = this.config.llm.options) == null ? void 0 : _v.hallucinationScoring) === true || this.config.llm.provider !== "ollama" && ((_w = this.config.llm.options) == null ? void 0 : _w.hallucinationScoring) !== false;
+        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;
         if (runHallucination) {
           hallucinationScoringPromise = scoreHallucination(this.llmProvider, fullReply, context).catch(() => void 0);
         }
@@ -7847,7 +7965,7 @@ ${context}`;
         const latency = {
           embedMs: Math.round(embedMs),
           retrieveMs: Math.round(retrieveMs),
-          rerankMs: ((_x = this.config.rag) == null ? void 0 : _x.useReranking) ? Math.round(rerankMs) : void 0,
+          rerankMs: arch !== "naive" && (arch === "retrieve-and-rerank" || ((_y = this.config.rag) == null ? void 0 : _y.useReranking)) ? Math.round(rerankMs) : void 0,
           generateMs: Math.round(generateMs),
           totalMs: Math.round(totalMs)
         };
@@ -7860,7 +7978,7 @@ ${context}`;
           totalTokens: promptTokens + completionTokens,
           estimatedCostUsd: estimateCostUsd(promptTokens, completionTokens, this.config.llm.model)
         };
-        const awaitHallucination = ((_y = this.config.llm.options) == null ? void 0 : _y.awaitHallucination) === true;
+        const awaitHallucination = ((_z = this.config.llm.options) == null ? void 0 : _z.awaitHallucination) === true;
         const [uiTransformation, hallucinationResult] = yield new __await(Promise.all([
           uiTransformationPromise,
           awaitHallucination ? hallucinationScoringPromise : Promise.resolve(void 0)
@@ -7889,7 +8007,7 @@ ${context}`;
           hallucinationReason: hScore.reason
         } : {});
         const trace = buildTrace(hallucinationResult);
-        if ((_z = this.config.telemetry) == null ? void 0 : _z.enabled) {
+        if ((_A = this.config.telemetry) == null ? void 0 : _A.enabled) {
           const telemetryUrl = this.config.telemetry.url || "/api/telemetry";
           const absoluteUrl = telemetryUrl.startsWith("http") ? telemetryUrl : (process.env.NEXT_PUBLIC_APP_URL || `http://localhost:${process.env.PORT || 3e3}`) + telemetryUrl;
           (async () => {
@@ -8165,6 +8283,7 @@ var Retrivora = class {
         this.config.projectId,
         (_a2 = this.config.vectorDb) == null ? void 0 : _a2.provider
       );
+      await LicenseVerifier.verifyIP(this.config.licenseKey);
     } catch (err) {
       throw wrapError(err, "CONFIGURATION_ERROR");
     }

package/package.json

@@ -1,9 +1,9 @@
 {
   "name": "@retrivora-ai/rag-engine",
-  "version": "1.9.8",
+  "version": "1.9.9",
   "description": "Retrivora AI is a plug-and-play AI engine for RAG chat experiences — generic vector DB + LLM provider, embeddable or standalone.",
   "author": "Abhinav Alkuchi",
-  "license": "MIT",
+  "license": "UNLICENSED",
   "keywords": [
     "rag",
     "retrieval-augmented-generation",
@@ -126,6 +126,7 @@
     "dotenv": "^17.4.2",
     "eslint": "^9",
     "eslint-config-next": "16.2.4",
+    "next": "16.2.10",
     "tailwindcss": "^4",
     "tsup": "^8.5.1",
     "typescript": "^5"

package/src/app/page.tsx

@@ -6,6 +6,7 @@ import { Navbar } from '@/components/Navbar';
 import { Hero } from '@/components/Hero';
 import { Lifecycle } from '@/components/Lifecycle';
 import { ArchitectureCardsSection } from '@/components/ArchitectureCardsSection';
+import { RagPatternsSection } from '@/components/RagPatternsSection';
 import { Documentation } from '@/components/Documentation';
 import { ChatWidget } from '@/components/ChatWidget';
 
@@ -17,6 +18,7 @@ export default function HomePage() {
       <div className="relative z-10 container mx-auto px-6 py-12">
         <Hero />
         <Lifecycle />
+        <RagPatternsSection />
         <ArchitectureCardsSection />
         <Documentation />
       </div>

package/src/components/constants.tsx

@@ -277,3 +277,227 @@ export const BORDER_RADIUS_MAP: Record<string, string> = {
   xl: 'rounded-xl',
   full: 'rounded-3xl',
 };
+
+export interface RagPatternStep {
+  label: string;
+  type: 'input' | 'process' | 'database' | 'model' | 'output';
+  color: string;
+}
+
+export interface RagPattern {
+  id: string;
+  name: string;
+  tagline: string;
+  description: string;
+  pros: string[];
+  cons: string[];
+  snippet: string;
+  color: string;
+  flowSteps: RagPatternStep[];
+}
+
+export const RAG_PATTERNS: RagPattern[] = [
+  {
+    id: 'naive',
+    name: 'Naive RAG',
+    tagline: 'Standard semantic search pipeline',
+    description: 'The foundation pattern: splits documents into text chunks, embeddings are stored in a vector database, and relevant context is fetched via direct query similarity to augment the model prompt.',
+    pros: ['Very low latency', 'Low resource cost', 'Simple implementation'],
+    cons: ['Struggles with complex multi-hop queries', 'No ranking optimization', 'Can return irrelevant context chunks'],
+    color: '#d97706', // Amber/Brown
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'naive-rag-demo',
+  vectorDatabase: {
+    provider: 'pinecone',
+    indexName: 'knowledge-base',
+  },
+  workflow: {
+    type: 'simple-rag'
+  }
+});`,
+    flowSteps: [
+      { label: 'Documents', type: 'input', color: 'bg-amber-100 text-amber-800 dark:bg-amber-950/40 dark:text-amber-300' },
+      { label: 'Chunking', type: 'process', color: 'bg-amber-50 text-amber-700 dark:bg-amber-900/20 dark:text-amber-400' },
+      { label: 'Embedding Model', type: 'model', color: 'bg-amber-100 text-amber-900 dark:bg-amber-800/30 dark:text-amber-200' },
+      { label: 'Vector Store', type: 'database', color: 'bg-amber-200 text-amber-950 dark:bg-amber-700/40 dark:text-amber-100' },
+      { label: 'Context Prompt', type: 'process', color: 'bg-amber-50 text-amber-700 dark:bg-amber-900/20 dark:text-amber-400' },
+      { label: 'LLM Generator', type: 'model', color: 'bg-indigo-100 text-indigo-900 dark:bg-indigo-800/30 dark:text-indigo-200' },
+    ]
+  },
+  {
+    id: 'retrieve-and-rerank',
+    name: 'Retrieve-and-rerank',
+    tagline: 'High-precision contextual relevance',
+    description: 'Enhances standard RAG by fetching a larger pool of vector candidates and feeding them through a cross-encoder Reranker model. The reranker scores context chunks dynamically relative to query intent, filtering noise.',
+    pros: ['Extremely high context precision', 'Better generation accuracy', 'Mitigates hallucination rates'],
+    cons: ['Adds slight latency bottleneck (~100-300ms)', 'Higher computational requirements'],
+    color: '#2563eb', // Blue
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'rerank-rag-demo',
+  vectorDatabase: {
+    provider: 'pinecone',
+    indexName: 'knowledge-base',
+  },
+  retrieval: {
+    strategy: 'contextual-compression',
+    useReranking: true, // Enables cross-encoder re-evaluation
+    topK: 5,
+  }
+});`,
+    flowSteps: [
+      { label: 'Query Input', type: 'input', color: 'bg-blue-100 text-blue-800 dark:bg-blue-950/40 dark:text-blue-300' },
+      { label: 'Vector Retrieval', type: 'database', color: 'bg-blue-200 text-blue-950 dark:bg-blue-700/40 dark:text-blue-100' },
+      { label: 'Reranker Model', type: 'model', color: 'bg-indigo-100 text-indigo-950 dark:bg-indigo-700/40 dark:text-indigo-100' },
+      { label: 'Filtered Context', type: 'process', color: 'bg-blue-50 text-blue-700 dark:bg-blue-900/20 dark:text-blue-400' },
+      { label: 'Generative Model', type: 'model', color: 'bg-violet-100 text-violet-900 dark:bg-violet-800/30 dark:text-violet-200' },
+    ]
+  },
+  {
+    id: 'multimodal',
+    name: 'Multimodal RAG',
+    tagline: 'Search and reason across text and media',
+    description: 'Ingests both unstructured text documents and visual/audio/structured assets. It queries multi-modal embeddings, feeding retrieved visuals and matched context fragments to multimodal LLMs (like Gemini or Claude) to support diagrams, tables, and media outputs.',
+    pros: ['Handles images, PDFs, charts, audio', 'Deeper context retrieval', 'Generates visual answers'],
+    cons: ['Multimodal embeddings are computationally expensive', 'Higher token payload sizes'],
+    color: '#7c3aed', // Purple
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'multimodal-rag-demo',
+  llm: {
+    provider: 'gemini',
+    model: 'gemini-2.5-flash', // Multimodal reasoning model
+  },
+  embedding: {
+    provider: 'gemini',
+    model: 'text-embedding-004',
+  },
+  workflow: {
+    type: 'multimodal-rag'
+  }
+});`,
+    flowSteps: [
+      { label: 'Multi-modal Docs', type: 'input', color: 'bg-purple-100 text-purple-800 dark:bg-purple-950/40 dark:text-purple-300' },
+      { label: 'Multi-modal Embed', type: 'model', color: 'bg-purple-50 text-purple-700 dark:bg-purple-900/20 dark:text-purple-400' },
+      { label: 'Vector Store', type: 'database', color: 'bg-purple-200 text-purple-950 dark:bg-purple-700/40 dark:text-purple-100' },
+      { label: 'Media Context', type: 'process', color: 'bg-purple-100 text-purple-950 dark:bg-purple-800/40 dark:text-purple-100' },
+      { label: 'Multimodal LLM', type: 'model', color: 'bg-violet-100 text-violet-900 dark:bg-violet-800/30 dark:text-violet-200' },
+    ]
+  },
+  {
+    id: 'graph',
+    name: 'Graph RAG',
+    tagline: 'Entity-relationship semantic reasoning',
+    description: 'Builds a knowledge graph structure from raw documents using an LLM extractor. Entity-to-entity and entity-to-document relationships are resolved, allowing queries to retrieve structured semantic contexts (nodes & relationships) to resolve structural, relational, and hierarchy-based questions.',
+    pros: ['Excellent for tracing connections', 'Resolves hierarchy and relationships', 'Grounded fact validation'],
+    cons: ['Heavy write-time pipeline ingestion cost', 'Requires graph database setup'],
+    color: '#059669', // Emerald
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'graph-rag-demo',
+  graphDb: {
+    provider: 'neo4j',
+    options: { uri: 'neo4j://localhost:7687' },
+  },
+  workflow: {
+    type: 'graph-rag'
+  }
+});`,
+    flowSteps: [
+      { label: 'Documents', type: 'input', color: 'bg-emerald-100 text-emerald-800 dark:bg-emerald-950/40 dark:text-emerald-300' },
+      { label: 'LLM Extractor', type: 'model', color: 'bg-emerald-50 text-emerald-700 dark:bg-emerald-900/20 dark:text-emerald-400' },
+      { label: 'Graph Database', type: 'database', color: 'bg-emerald-200 text-emerald-950 dark:bg-emerald-700/40 dark:text-emerald-100' },
+      { label: 'Relation Search', type: 'process', color: 'bg-emerald-100 text-emerald-950 dark:bg-emerald-800/40 dark:text-emerald-100' },
+      { label: 'LLM Reasoner', type: 'model', color: 'bg-indigo-100 text-indigo-900 dark:bg-indigo-800/30 dark:text-indigo-200' },
+    ]
+  },
+  {
+    id: 'hybrid',
+    name: 'Hybrid RAG',
+    tagline: 'Best of both worlds: Vector + Graph',
+    description: 'Queries both a vector database for semantic match proximity and a graph database for topological linkages. It joins vector search results and sub-graph entity contexts into a unified prompt, resolving complex interconnected schemas.',
+    pros: ['Unparalleled query coverage', 'Combines semantic proximity and graph relationships', 'Robust grounding'],
+    cons: ['Most complex RAG configuration', 'Double database query overhead'],
+    color: '#d97706', // Orange
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'hybrid-rag-demo',
+  vectorDatabase: { provider: 'mongodb', indexName: 'default' },
+  graphDb: { provider: 'neo4j', options: { uri: 'neo4j://...' } },
+  workflow: {
+    type: 'hybrid-rag'
+  }
+});`,
+    flowSteps: [
+      { label: 'Query', type: 'input', color: 'bg-amber-100 text-amber-800 dark:bg-amber-950/40 dark:text-amber-300' },
+      { label: 'Vector Store', type: 'database', color: 'bg-orange-100 text-orange-950 dark:bg-orange-700/40 dark:text-orange-100' },
+      { label: 'Graph Database', type: 'database', color: 'bg-emerald-100 text-emerald-950 dark:bg-emerald-700/40 dark:text-emerald-100' },
+      { label: 'Unified Context', type: 'process', color: 'bg-amber-50 text-amber-700 dark:bg-amber-900/20 dark:text-amber-400' },
+      { label: 'LLM Reasoner', type: 'model', color: 'bg-indigo-100 text-indigo-900 dark:bg-indigo-800/30 dark:text-indigo-200' },
+    ]
+  },
+  {
+    id: 'agentic-router',
+    name: 'Agentic RAG (Router)',
+    tagline: 'Dynamic intent-based query routing',
+    description: 'Uses an intelligent fast router agent to analyze the intent of the incoming query. The router dynamically decides whether to fetch data from the Vector DB, query a Graph DB, fallback to web search, or trigger external API tools.',
+    pros: ['Highly adaptive workflow', 'Saves tokens on simple queries', 'Automatically invokes external API tools'],
+    cons: ['Adds slight routing classification latency', 'Dependent on router accuracy'],
+    color: '#0891b2', // Cyan
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'agentic-router-demo',
+  workflow: {
+    type: 'agentic-router'
+  }
+});`,
+    flowSteps: [
+      { label: 'User Query', type: 'input', color: 'bg-cyan-100 text-cyan-800 dark:bg-cyan-950/40 dark:text-cyan-300' },
+      { label: 'Router Agent', type: 'model', color: 'bg-indigo-100 text-indigo-900 dark:bg-indigo-800/30 dark:text-indigo-200' },
+      { label: 'Vector / Graph / Tools', type: 'database', color: 'bg-cyan-200 text-cyan-950 dark:bg-cyan-700/40 dark:text-cyan-100' },
+      { label: 'Selected Context', type: 'process', color: 'bg-cyan-50 text-cyan-700 dark:bg-cyan-900/20 dark:text-cyan-400' },
+      { label: 'LLM Generator', type: 'model', color: 'bg-violet-100 text-violet-900 dark:bg-violet-800/30 dark:text-violet-200' },
+    ]
+  },
+  {
+    id: 'multi-agent',
+    name: 'Agent RAG (Multi-Agent)',
+    tagline: 'Collaborative swarm-based problem solving',
+    description: 'Deploys a supervisor agent that coordinates a group of specialized worker agents (e.g., search agent, tool executor agent, web crawler agent). Agents collaborate, pass information, check each other\'s facts, and compile a final answer.',
+    pros: ['Excellent for complex research tasks', 'Self-correcting query pipelines', 'Multi-step action sequences'],
+    cons: ['High latency (requires multiple LLM cycles)', 'Higher cost per query', 'Complex state synchronization'],
+    color: '#ec4899', // Pink
+    snippet: `import { Retrivora } from '@retrivora-ai/rag-engine/server';
+
+const ai = new Retrivora({
+  projectId: 'multi-agent-demo',
+  workflow: {
+    type: 'multi-agent'
+  },
+  mcpServers: [
+    {
+      name: 'search-engine',
+      transport: 'stdio',
+      command: 'npx',
+      args: ['-y', '@modelcontextprotocol/server-web-search']
+    }
+  ]
+});`,
+    flowSteps: [
+      { label: 'Goal Query', type: 'input', color: 'bg-pink-100 text-pink-800 dark:bg-pink-950/40 dark:text-pink-300' },
+      { label: 'Supervisor Agent', type: 'model', color: 'bg-pink-200 text-pink-950 dark:bg-pink-800/30 dark:text-pink-100' },
+      { label: 'Specialist Agents', type: 'model', color: 'bg-indigo-100 text-indigo-900 dark:bg-indigo-900/30 dark:text-indigo-200' },
+      { label: 'Tools / Vector DBs', type: 'database', color: 'bg-pink-200 text-pink-950 dark:bg-pink-700/40 dark:text-pink-100' },
+      { label: 'Consensus Context', type: 'process', color: 'bg-pink-50 text-pink-700 dark:bg-pink-900/20 dark:text-pink-400' },
+      { label: 'Final Output', type: 'output', color: 'bg-indigo-600 text-white shadow-sm' },
+    ]
+  }
+];
+

package/src/config/RagConfig.ts

@@ -182,7 +182,7 @@ export interface RAGConfig {
   /** Characters used to split text, in order of priority */
   separators?: string[];
   /** Overall RAG architecture pattern */
-  architecture?: 'simple' | 'graph' | 'hybrid' | 'agentic';
+  architecture?: 'simple' | 'graph' | 'hybrid' | 'agentic' | 'naive' | 'retrieve-and-rerank' | 'multimodal' | 'agentic-router' | 'multi-agent';
   /** Chunking strategy to use during ingestion */
   chunkingStrategy?: 'recursive' | 'markdown' | 'semantic' | 'llamaindex';
   /** Whether to use query transformation (e.g. HyDE) */
@@ -249,7 +249,7 @@ export interface RetrievalConfig {
 
 export interface WorkflowConfig {
   /** High-level workflow selector from the SDK prompt. */
-  type?: 'simple-rag' | 'rag' | 'hybrid-rag' | 'graph-rag' | 'agentic-rag' | 'agentic';
+  type?: 'simple-rag' | 'rag' | 'hybrid-rag' | 'graph-rag' | 'agentic-rag' | 'agentic' | 'naive-rag' | 'retrieve-and-rerank' | 'multimodal-rag' | 'agentic-router' | 'multi-agent-rag' | 'multi-agent';
   /** Future workflow/provider-specific options. */
   options?: Record<string, unknown>;
 }