diff --git a/src/partnerRecruiter.ts b/src/partnerRecruiter.ts
new file mode 100644
index 0000000..2ac2c20
--- /dev/null
+++ b/src/partnerRecruiter.ts
@@ -0,0 +1,56 @@
+// partnerRecruiter.ts
+
+/**
+ * PartnerRecruiterAgent – simple utility to discover potential software partners in Medellín.
+ * In a production system this would call LinkedIn, Crunchbase, or a local DB. Here we provide
+ * a deterministic mock that can be extended later.
+ */
+
+export interface CompanyProfile {
+  name: string;
+  sector: string;
+  website: string;
+  employeeCount: number;
+  description: string;
+}
+
+/**
+ * Returns a curated list of software companies in Medellín that match the given sector filter.
+ * @param sector Desired sector (e.g., "AI", "FinTech", "DevOps"). If omitted, returns all.
+ */
+export function findMedellinPartners(sector?: string): CompanyProfile[] {
+  const companies: CompanyProfile[] = [
+    {
+      name: "CleverTech",
+      sector: "AI",
+      website: "https://clevertech.co",
+      employeeCount: 45,
+      description: "AI‑driven automation platform for enterprises.",
+    },
+    {
+      name: "FinScope",
+      sector: "FinTech",
+      website: "https://finscope.com",
+      employeeCount: 30,
+      description: "Analytics suite for financial institutions.",
+    },
+    {
+      name: "DevOpsM",
+      sector: "DevOps",
+      website: "https://devopsm.io",
+      employeeCount: 22,
+      description: "Continuous delivery pipelines with zero‑downtime deployments.",
+    },
+    {
+      name: "DataPulse",
+      sector: "Data",
+      website: "https://datapulse.io",
+      employeeCount: 55,
+      description: "Real‑time data streaming and observability.",
+    },
+  ];
+
+  if (!sector) return companies;
+  const lowered = sector.toLowerCase();
+  return companies.filter((c) => c.sector.toLowerCase().includes(lowered));
+}
diff --git a/src/rag.ts b/src/rag.ts
new file mode 100644
index 0000000..aa51605
--- /dev/null
+++ b/src/rag.ts
@@ -0,0 +1,156 @@
+// rag.ts
+
+/**
+ * Enhanced Retrieval-Augmented Generation (RAG) utilities for Isaac Forge.
+ * Implements document ingestion, external scholarly metadata fetching, vector indexing,
+ * and citation-aware retrieval.
+ */
+
+import axios from 'axios';
+import { Document, VectorStoreIndex, SimpleRetriever } from 'llamaindex'; // Assume llamaindex provides these exports
+import { AgentFunction, FunctionDescriptor } from './types';
+import { logDebugMessage } from './util';
+import { v4 as uuidv4 } from 'uuid';
+
+/** Configuration for external services */
+export interface RAGConfig {
+  /** Semantic Scholar API base URL */
+  semanticScholarBase?: string;
+  /** Optional API key for Semantic Scholar (if needed) */
+  semanticScholarKey?: string;
+  /** LlamaIndex index persistence directory */
+  indexPath?: string;
+}
+
+/** Default configuration values */
+const DEFAULT_CONFIG: Required<RAGConfig> = {
+  semanticScholarBase: 'https://api.semanticscholar.org/graph/v1/paper/search',
+  semanticScholarKey: '',
+  indexPath: './rag-index',
+};
+
+/** Ingest a raw text document and return a LlamaIndex Document */
+export function ingestDocument(content: string, metadata: Record<string, any> = {}): Document {
+  const id = metadata.id ?? uuidv4();
+  const doc = new Document({
+    id,
+    text: content,
+    metadata: { ...metadata, source: metadata.source ?? 'local' },
+  });
+  logDebugMessage(true, 'Ingested document', { id, length: content.length, metadata });
+  return doc;
+}
+
+/** Fetch scholarly papers from Semantic Scholar matching a query */
+export async function fetchSemanticScholarPapers(
+  query: string,
+  limit: number = 5,
+  config?: RAGConfig,
+): Promise<Document[]> {
+  const cfg = { ...DEFAULT_CONFIG, ...config };
+  const params = new URLSearchParams({
+    query,
+    limit: limit.toString(),
+    fields: 'title,abstract,year,authors,venue,url',
+  });
+  const url = `${cfg.semanticScholarBase}?${params.toString()}`;
+  const headers: Record<string, string> = {};
+  if (cfg.semanticScholarKey) {
+    headers['x-api-key'] = cfg.semanticScholarKey;
+  }
+  try {
+    const resp = await axios.get(url, { headers });
+    const papers = resp.data?.data ?? [];
+    return papers.map((p: any) => {
+      const text = `${p.title}\n${p.abstract ?? ''}`;
+      const meta = {
+        id: p.paperId,
+        source: 'semantic-scholar',
+        title: p.title,
+        year: p.year,
+        authors: p.authors?.map((a: any) => a.name).join(', '),
+        venue: p.venue,
+        url: p.url,
+      };
+      return ingestDocument(text, meta);
+    });
+  } catch (e: any) {
+    logDebugMessage(true, 'Semantic Scholar fetch error', { error: e.message });
+    throw new Error(`Failed to fetch from Semantic Scholar: ${e.message}`);
+  }
+}
+
+/** Build or load a vector index from a collection of documents */
+export async function buildIndex(
+  docs: Document[],
+  config?: RAGConfig,
+): Promise<VectorStoreIndex> {
+  const cfg = { ...DEFAULT_CONFIG, ...config };
+  // LlamaIndex can persist to a directory; we assume a simple in‑memory index for now.
+  const index = await VectorStoreIndex.fromDocuments(docs);
+  logDebugMessage(true, 'Built vector index', { docCount: docs.length, indexPath: cfg.indexPath });
+  return index;
+}
+
+/** Retrieve top‑k documents relevant to a query */
+export async function retrieveRelevantDocs(
+  query: string,
+  index: VectorStoreIndex,
+  topK: number = 4,
+): Promise<Document[]> {
+  const retriever = new SimpleRetriever({
+    index,
+    similarityTopK: topK,
+  });
+  const results = await retriever.retrieve(query);
+  logDebugMessage(true, 'Retrieved relevant docs', { query, topK, resultCount: results.length });
+  return results;
+}
+
+/** Agent function wrapper exposing RAG retrieval to Swarm agents */
+export const retrieveRelevantDocsFunction: AgentFunction = {
+  name: 'retrieveRelevantDocs',
+  func: async (args: { query: string; topK?: number }) => {
+    const { query, topK } = args;
+    // For demonstration we lazily build an index from a static corpus. In production this
+    // would be cached or persisted.
+    const localDocs = [
+      ingestDocument('Placeholder local scientific note about quantum entanglement.', {
+        source: 'local',
+        title: 'Quantum Entanglement Overview',
+      }),
+    ];
+    const scholarlyDocs = await fetchSemanticScholarPapers(query, topK ?? 4);
+    const allDocs = [...localDocs, ...scholarlyDocs];
+    const index = await buildIndex(allDocs);
+    const relevant = await retrieveRelevantDocs(query, index, topK ?? 4);
+    // Return a formatted string that includes citations.
+    const formatted = relevant
+      .map((doc, i) => {
+        const meta = doc.metadata as any;
+        const citation = meta.source === 'semantic-scholar' ? `[${meta.year}] ${meta.title}` : `[Local] ${meta.title ?? 'Document']`;
+        return `Result ${i + 1}: ${doc.text.slice(0, 200)}...\nCitation: ${citation}`;
+      })
+      .join('\n\n');
+    return formatted;
+  },
+  descriptor: {
+    name: 'retrieveRelevantDocs',
+    description: 'Retrieve top‑k relevant scientific documents (local or Semantic Scholar) for a given query, returning a formatted string with citations.',
+    parameters: {
+      query: {
+        type: 'string',
+        required: true,
+        description: 'Natural language query describing the research question.',
+      },
+      topK: {
+        type: 'number',
+        required: false,
+        description: 'Number of documents to retrieve (default 4).',
+      },
+    },
+  },
+};
+
+/** Export a collection of RAG‑related functions for easy registration */
+export const RAG_FUNCTIONS: AgentFunction[] = [retrieveRelevantDocsFunction];