@memvid/sdk 2.0.156 → 2.0.157

package/dist/documents/index.d.ts

@@ -17,7 +17,8 @@
 import type { ParseOptions, ParseResult } from "./types";
 export * from "./types";
 export { parsePdf } from "./pdf";
-export { parseXlsx } from "./xlsx";
+export { parseXlsx, parseXlsxStructured } from "./xlsx";
+export type { XlsxStructuredChunk, XlsxDetectedTable, XlsxStructuredOptions, XlsxStructuredResult, } from "./xlsx";
 export { parsePptx } from "./pptx";
 export { parseDocx } from "./docx";
 /**

package/dist/documents/index.js

@@ -30,7 +30,7 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
     for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.parseDocx = exports.parsePptx = exports.parseXlsx = exports.parsePdf = void 0;
+exports.parseDocx = exports.parsePptx = exports.parseXlsxStructured = exports.parseXlsx = exports.parsePdf = void 0;
 exports.parse = parse;
 exports.isSupportedFormat = isSupportedFormat;
 exports.getDocumentType = getDocumentType;
@@ -45,6 +45,7 @@ var pdf_2 = require("./pdf");
 Object.defineProperty(exports, "parsePdf", { enumerable: true, get: function () { return pdf_2.parsePdf; } });
 var xlsx_2 = require("./xlsx");
 Object.defineProperty(exports, "parseXlsx", { enumerable: true, get: function () { return xlsx_2.parseXlsx; } });
+Object.defineProperty(exports, "parseXlsxStructured", { enumerable: true, get: function () { return xlsx_2.parseXlsxStructured; } });
 var pptx_2 = require("./pptx");
 Object.defineProperty(exports, "parsePptx", { enumerable: true, get: function () { return pptx_2.parsePptx; } });
 var docx_2 = require("./docx");

package/dist/documents/xlsx.d.ts

@@ -1,6 +1,82 @@
 /**
  * Excel Parser with error handling
- * Uses exceljs - no Rust fallback available for XLSX
+ *
+ * Two strategies:
+ * - `parseXlsx()` — JS-based via exceljs (legacy, per-sheet flat text)
+ * - `parseXlsxStructured()` — Rust-native via memvid-core (high accuracy,
+ *   header-value pairing, merged cell support, semantic chunking)
  */
 import type { ParseOptions, ParseResult } from "./types";
 export declare function parseXlsx(filePath: string, options?: ParseOptions): Promise<ParseResult>;
+/** A single structured chunk from the XLSX extraction pipeline. */
+export interface XlsxStructuredChunk {
+    /** Chunk text with [Sheet:] prefix and Header: Value formatting */
+    text: string;
+    /** Chunk type: "Table", "TableContinuation", "Text", etc. */
+    chunkType: string;
+    /** 0-based chunk index */
+    index: number;
+    /** Table/element identifier */
+    elementId?: string;
+    /** Header context for continuation chunks */
+    context?: string;
+}
+/** A detected table with headers and metadata. */
+export interface XlsxDetectedTable {
+    name: string;
+    sheetName: string;
+    headers: string[];
+    headerRow: number;
+    firstDataRow: number;
+    lastDataRow: number;
+    firstCol: number;
+    lastCol: number;
+    confidence: number;
+    columnTypes: string[];
+}
+/** Options for structured XLSX extraction. */
+export interface XlsxStructuredOptions {
+    /** Target chunk size in characters (default: 1200) */
+    maxChars?: number;
+    /** Maximum chunks to produce (default: 500) */
+    maxChunks?: number;
+}
+/** Result of structured XLSX extraction. */
+export interface XlsxStructuredResult {
+    /** Backward-compatible flat text */
+    text: string;
+    /** Semantic chunks with header-value pairing */
+    chunks: XlsxStructuredChunk[];
+    /** Detected tables with metadata */
+    tables: XlsxDetectedTable[];
+    /** Extraction diagnostics */
+    diagnostics: {
+        warnings: string[];
+        tablesProcessed: number;
+        tablesSplit: number;
+    };
+    /** Extraction time in milliseconds */
+    timingMs: number;
+}
+/**
+ * Parse an XLSX file using the Rust structured extraction pipeline.
+ *
+ * This provides much higher search accuracy than `parseXlsx()` by:
+ * - Detecting table boundaries and headers automatically
+ * - Formatting rows as `Header: Value | Header: Value` pairs
+ * - Propagating merged cells
+ * - Detecting number formats (dates, currency, percentages)
+ * - Never splitting rows across chunk boundaries
+ *
+ * @example
+ * ```typescript
+ * const result = await parseXlsxStructured("./proforma.xlsx");
+ * console.log(`${result.tables.length} tables, ${result.chunks.length} chunks`);
+ *
+ * // Ingest chunks into memvid for high-accuracy search
+ * for (const chunk of result.chunks) {
+ *   await mem.put({ text: chunk.text, title: `XLSX chunk ${chunk.index}` });
+ * }
+ * ```
+ */
+export declare function parseXlsxStructured(filePath: string, options?: XlsxStructuredOptions): Promise<XlsxStructuredResult>;

package/dist/documents/xlsx.js

@@ -1,7 +1,11 @@
 "use strict";
 /**
  * Excel Parser with error handling
- * Uses exceljs - no Rust fallback available for XLSX
+ *
+ * Two strategies:
+ * - `parseXlsx()` — JS-based via exceljs (legacy, per-sheet flat text)
+ * - `parseXlsxStructured()` — Rust-native via memvid-core (high accuracy,
+ *   header-value pairing, merged cell support, semantic chunking)
  */
 var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
     if (k2 === undefined) k2 = k;
@@ -38,6 +42,7 @@ var __importStar = (this && this.__importStar) || (function () {
 })();
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.parseXlsx = parseXlsx;
+exports.parseXlsxStructured = parseXlsxStructured;
 /** Extract the display value from an exceljs cell */
 function getCellValue(cellValue) {
     if (cellValue == null || cellValue === "")
@@ -115,3 +120,57 @@ async function parseXlsx(filePath, options) {
             `Ensure the file is a valid .xlsx/.xls file.`);
     }
 }
+/**
+ * Parse an XLSX file using the Rust structured extraction pipeline.
+ *
+ * This provides much higher search accuracy than `parseXlsx()` by:
+ * - Detecting table boundaries and headers automatically
+ * - Formatting rows as `Header: Value | Header: Value` pairs
+ * - Propagating merged cells
+ * - Detecting number formats (dates, currency, percentages)
+ * - Never splitting rows across chunk boundaries
+ *
+ * @example
+ * ```typescript
+ * const result = await parseXlsxStructured("./proforma.xlsx");
+ * console.log(`${result.tables.length} tables, ${result.chunks.length} chunks`);
+ *
+ * // Ingest chunks into memvid for high-accuracy search
+ * for (const chunk of result.chunks) {
+ *   await mem.put({ text: chunk.text, title: `XLSX chunk ${chunk.index}` });
+ * }
+ * ```
+ */
+async function parseXlsxStructured(filePath, options) {
+    try {
+        // Try native Rust extraction first
+        const native = require("../../index.js");
+        if (typeof native.parseXlsxStructured === "function") {
+            const nativeOpts = options
+                ? { maxChars: options.maxChars, maxChunks: options.maxChunks }
+                : undefined;
+            return await native.parseXlsxStructured(filePath, nativeOpts);
+        }
+    }
+    catch {
+        // Native not available — fall through to JS fallback
+    }
+    // Fallback: use parseXlsx and wrap in structured format
+    const legacy = await parseXlsx(filePath);
+    const chunks = legacy.items.map((item, i) => ({
+        text: `[Sheet: ${item.name || `Sheet${item.number}`}]\n${item.text}`,
+        chunkType: "Table",
+        index: i,
+    }));
+    return {
+        text: legacy.items.map((item) => item.text).join("\n\n"),
+        chunks,
+        tables: [],
+        diagnostics: {
+            warnings: ["Native structured extraction unavailable; using JS fallback"],
+            tablesProcessed: 0,
+            tablesSplit: 0,
+        },
+        timingMs: 0,
+    };
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@memvid/sdk",
-  "version": "2.0.156",
+  "version": "2.0.157",
   "description": "Single-file AI memory system for Node.js. Store, search, and query documents with built-in RAG.",
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
@@ -41,11 +41,11 @@
     "node": ">=18"
   },
   "optionalDependencies": {
-    "@memvid/sdk-darwin-arm64": "2.0.156",
-    "@memvid/sdk-darwin-x64": "2.0.156",
-    "@memvid/sdk-linux-x64-gnu": "2.0.156",
-    "@memvid/sdk-linux-arm64-gnu": "2.0.156",
-    "@memvid/sdk-win32-x64-msvc": "2.0.156"
+    "@memvid/sdk-darwin-arm64": "2.0.157",
+    "@memvid/sdk-darwin-x64": "2.0.157",
+    "@memvid/sdk-linux-x64-gnu": "2.0.157",
+    "@memvid/sdk-linux-arm64-gnu": "2.0.157",
+    "@memvid/sdk-win32-x64-msvc": "2.0.157"
   },
   "peerDependencies": {
     "@langchain/core": ">=0.3.0",
@@ -89,7 +89,6 @@
     "langchain": ">=0.3.0",
     "llamaindex": ">=0.12.0",
     "officeparser": "^6.0.2",
-    "unpdf": "^1.4.0",
-    "xlsx": "^0.18.5"
+    "unpdf": "^1.4.0"
   }
 }