@alexion42/pi-web-search 0.1.0

package/package.json

@@ -0,0 +1,46 @@
+{
+  "name": "@alexion42/pi-web-search",
+  "version": "0.1.0",
+  "description": "Exa-powered web search and content extraction for Pi coding agent",
+  "type": "module",
+  "scripts": {
+    "test": "node --test"
+  },
+  "keywords": [
+    "pi-package",
+    "pi",
+    "pi-coding-agent",
+    "extension",
+    "web-search",
+    "exa",
+    "fetch",
+    "scraping"
+  ],
+  "author": "Alexion Fortytwo <alexion@lexiupon.com>",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/nicobailon/pi-web-access.git"
+  },
+  "bugs": {
+    "url": "https://github.com/nicobailon/pi-web-access/issues"
+  },
+  "homepage": "https://github.com/nicobailon/pi-web-access#readme",
+  "dependencies": {
+    "@mozilla/readability": "^0.5.0",
+    "linkedom": "^0.16.0",
+    "p-limit": "^6.1.0",
+    "turndown": "^7.2.0",
+    "unpdf": "^1.6.2"
+  },
+  "peerDependencies": {
+    "@mariozechner/pi-ai": ">=0.6.0",
+    "@mariozechner/pi-coding-agent": ">=0.37.3",
+    "@mariozechner/pi-tui": ">=0.37.3"
+  },
+  "pi": {
+    "extensions": [
+      "./index.ts"
+    ]
+  }
+}

package/pdf-extract.ts

@@ -0,0 +1,192 @@
+/**
+ * PDF Content Extractor
+ * 
+ * Extracts text from PDF files and saves to markdown.
+ * Uses unpdf (pdfjs-dist wrapper) for text extraction.
+ */
+
+import { getDocumentProxy } from "unpdf";
+import { writeFile, mkdir } from "node:fs/promises";
+import { join, basename } from "node:path";
+import { homedir } from "node:os";
+
+export interface PDFExtractResult {
+  title: string;
+  pages: number;
+  chars: number;
+  outputPath: string;
+}
+
+export interface PDFExtractOptions {
+  maxPages?: number;
+  outputDir?: string;
+  filename?: string;
+}
+
+const DEFAULT_MAX_PAGES = 100;
+const DEFAULT_OUTPUT_DIR = join(homedir(), "Downloads");
+
+/**
+ * Extract text from a PDF buffer and save to markdown file
+ */
+export async function extractPDFToMarkdown(
+  buffer: ArrayBuffer,
+  url: string,
+  options: PDFExtractOptions = {}
+): Promise<PDFExtractResult> {
+  const {
+    maxPages = DEFAULT_MAX_PAGES,
+    outputDir = DEFAULT_OUTPUT_DIR,
+    filename,
+  } = options;
+
+  const safeMaxPages = Number.isFinite(maxPages)
+    ? Math.max(1, Math.floor(maxPages))
+    : DEFAULT_MAX_PAGES;
+
+  const pdf = await getDocumentProxy(new Uint8Array(buffer));
+  const metadata = await pdf.getMetadata();
+  const metadataInfo = metadata.info && typeof metadata.info === "object"
+    ? metadata.info as Record<string, unknown>
+    : null;
+
+  // Extract title from metadata or URL
+  const metaTitle = typeof metadataInfo?.Title === "string" ? metadataInfo.Title : undefined;
+  const metaAuthor = typeof metadataInfo?.Author === "string" ? metadataInfo.Author : undefined;
+  const urlTitle = extractTitleFromURL(url);
+  const title = metaTitle?.trim() || urlTitle;
+
+  // Determine pages to extract
+  const pagesToExtract = Math.min(pdf.numPages, safeMaxPages);
+  const truncated = pdf.numPages > safeMaxPages;
+
+  // Extract text page by page for better structure
+  const pages: { pageNum: number; text: string }[] = [];
+  for (let i = 1; i <= pagesToExtract; i++) {
+    const page = await pdf.getPage(i);
+    const textContent = await page.getTextContent();
+    const pageText = textContent.items
+      .map((item: unknown) => {
+        const textItem = item as { str?: string };
+        return textItem.str || "";
+      })
+      .join(" ")
+      .replace(/\s+/g, " ")
+      .trim();
+    
+    if (pageText) {
+      pages.push({ pageNum: i, text: pageText });
+    }
+  }
+
+  // Build markdown content
+  const lines: string[] = [];
+  
+  // Header with metadata
+  lines.push(`# ${title}`);
+  lines.push("");
+  lines.push(`> Source: ${url}`);
+  lines.push(`> Pages: ${pdf.numPages}${truncated ? ` (extracted first ${pagesToExtract})` : ""}`);
+  if (metaAuthor) {
+    lines.push(`> Author: ${metaAuthor}`);
+  }
+  lines.push("");
+  lines.push("---");
+  lines.push("");
+
+  // Content with page markers
+  for (let i = 0; i < pages.length; i++) {
+    if (i > 0) {
+      lines.push("");
+      lines.push(`<!-- Page ${pages[i].pageNum} -->`);
+      lines.push("");
+    }
+    lines.push(pages[i].text);
+  }
+
+  if (truncated) {
+    lines.push("");
+    lines.push("---");
+    lines.push("");
+    lines.push(`*[Truncated: Only first ${pagesToExtract} of ${pdf.numPages} pages extracted]*`);
+  }
+
+  const content = lines.join("\n");
+
+  // Generate output filename
+  const outputFilename = filename || sanitizeFilename(title) + ".md";
+  const outputPath = join(outputDir, outputFilename);
+
+  // Ensure output directory exists
+  await mkdir(outputDir, { recursive: true });
+
+  // Write file
+  await writeFile(outputPath, content, "utf-8");
+
+  return {
+    title,
+    pages: pdf.numPages,
+    chars: content.length,
+    outputPath,
+  };
+}
+
+/**
+ * Extract a reasonable title from URL
+ */
+function extractTitleFromURL(url: string): string {
+  try {
+    const urlObj = new URL(url);
+    const pathname = urlObj.pathname;
+    
+    // Get filename without extension
+    let filename = basename(pathname, ".pdf");
+    
+    // Handle arxiv URLs: /pdf/1706.03762 → "arxiv-1706.03762"
+    if (urlObj.hostname.includes("arxiv.org")) {
+      const match = pathname.match(/\/(?:pdf|abs)\/(\d+\.\d+)/);
+      if (match) {
+        filename = `arxiv-${match[1]}`;
+      }
+    }
+    
+    // Clean up filename
+    filename = filename
+      .replace(/[_-]+/g, " ")
+      .replace(/\s+/g, " ")
+      .trim();
+    
+    return filename || "document";
+  } catch {
+    return "document";
+  }
+}
+
+/**
+ * Sanitize string for use as filename
+ */
+function sanitizeFilename(name: string): string {
+  return name
+    .toLowerCase()
+    .replace(/[^a-z0-9\s-]/g, "")
+    .replace(/\s+/g, "-")
+    .replace(/-+/g, "-")
+    .slice(0, 100)
+    .replace(/^-|-$/g, "")
+    || "document";
+}
+
+/**
+ * Check if URL or content-type indicates a PDF
+ */
+export function isPDF(url: string, contentType?: string): boolean {
+  if (contentType?.includes("application/pdf")) {
+    return true;
+  }
+  try {
+    const urlObj = new URL(url);
+    return urlObj.pathname.toLowerCase().endsWith(".pdf");
+  } catch {
+    return false;
+  }
+}

package/rsc-extract.ts

@@ -0,0 +1,338 @@
+/**
+ * RSC Content Extractor
+ * 
+ * Extracts readable content from Next.js React Server Components (RSC) flight payloads.
+ * RSC pages embed content as JSON in <script>self.__next_f.push([...])</script> tags.
+ */
+
+export interface RSCExtractResult {
+  title: string;
+  content: string;
+}
+
+export function extractRSCContent(html: string): RSCExtractResult | null {
+  if (!html.includes("self.__next_f.push")) {
+    return null;
+  }
+
+  // Parse all RSC chunks into a map
+  const chunkMap = new Map<string, string>();
+  const scriptRegex = /<script>self\.__next_f\.push\(\[1,"([\s\S]*?)"\]\)<\/script>/g;
+
+  for (const match of html.matchAll(scriptRegex)) {
+    let content: string;
+    try {
+      content = JSON.parse('"' + match[1] + '"');
+    } catch {
+      continue;
+    }
+
+    // Parse each line as "id:payload"
+    // Lines are separated by \n, each line is one chunk
+    // Chunk IDs are hex strings, typically 1-4 chars (supports up to 65535 chunks)
+    for (const line of content.split("\n")) {
+      if (!line.trim()) continue;
+      
+      const colonIdx = line.indexOf(":");
+      if (colonIdx <= 0 || colonIdx > 4) continue;
+      
+      const id = line.slice(0, colonIdx);
+      if (!/^[0-9a-f]+$/i.test(id)) continue;
+      
+      const payload = line.slice(colonIdx + 1);
+      if (!payload) continue;
+      
+      const existing = chunkMap.get(id);
+      if (!existing || payload.length > existing.length) {
+        chunkMap.set(id, payload);
+      }
+    }
+  }
+
+  if (chunkMap.size === 0) return null;
+
+  // Extract title
+  const titleMatch = html.match(/<title[^>]*>([^<]+)<\/title>/);
+  const title = titleMatch?.[1]?.split("|")[0]?.trim() || "";
+
+  // Parse and cache parsed chunks
+  const parsedCache = new Map<string, unknown>();
+  
+  function getParsedChunk(id: string): unknown | null {
+    if (parsedCache.has(id)) return parsedCache.get(id);
+    
+    const chunk = chunkMap.get(id);
+    if (!chunk || !chunk.startsWith("[")) {
+      parsedCache.set(id, null);
+      return null;
+    }
+    
+    try {
+      const parsed = JSON.parse(chunk);
+      parsedCache.set(id, parsed);
+      return parsed;
+    } catch {
+      parsedCache.set(id, null);
+      return null;
+    }
+  }
+
+  // Extract markdown from nodes, resolving refs on the fly
+  type Node = unknown;
+  const visitedRefs = new Set<string>();
+
+  function extractNode(node: Node, ctx = { inTable: false, inCode: false }): string {
+    if (node === null || node === undefined) return "";
+    
+    if (typeof node === "string") {
+      // Check if it's a reference like "$L30"
+      const refMatch = node.match(/^\$L([0-9a-f]+)$/i);
+      if (refMatch) {
+        const refId = refMatch[1];
+        if (visitedRefs.has(refId)) return ""; // Prevent cycles
+        visitedRefs.add(refId);
+        const refNode = getParsedChunk(refId);
+        const result = refNode ? extractNode(refNode, ctx) : "";
+        visitedRefs.delete(refId);
+        return result;
+      }
+      // Filter out RSC-specific artifacts, but preserve content inside code blocks
+      if (!ctx.inCode && (node === "$undefined" || node === "$" || /^\$[A-Z]/.test(node))) return "";
+      return node.trim() ? node : "";
+    }
+    
+    if (typeof node === "number") return String(node);
+    if (typeof node === "boolean") return "";
+    if (!Array.isArray(node)) return "";
+
+    // RSC element: ["$", "tag", key, props]
+    if (node[0] === "$" && typeof node[1] === "string") {
+      const tag = node[1] as string;
+      const props = (node[3] || {}) as Record<string, unknown>;
+
+      // Skip non-content
+      const skipTags = ["script", "style", "svg", "path", "circle", "link", "meta", 
+                        "template", "button", "input", "nav", "footer", "aside"];
+      if (skipTags.includes(tag)) return "";
+
+      // Component ref like $L25
+      if (tag.startsWith("$L")) {
+        const refId = tag.slice(2);
+        if (visitedRefs.has(refId)) return "";
+        
+        // Check for heading components with baseId
+        if (props.baseId && props.children) {
+          return `## ${String(props.children)}\n\n`;
+        }
+        
+        visitedRefs.add(refId);
+        const refNode = getParsedChunk(refId);
+        let result = "";
+        if (refNode) {
+          result = extractNode(refNode, ctx);
+        } else if (props.children) {
+          result = extractNode(props.children as Node, ctx);
+        }
+        visitedRefs.delete(refId);
+        return result;
+      }
+
+      const children = props.children;
+      const content = children ? extractNode(children as Node, ctx) : "";
+
+      switch (tag) {
+        case "h1": return `# ${content.trim()}\n\n`;
+        case "h2": return `## ${content.trim()}\n\n`;
+        case "h3": return `### ${content.trim()}\n\n`;
+        case "h4": return `#### ${content.trim()}\n\n`;
+        case "h5": return `##### ${content.trim()}\n\n`;
+        case "h6": return `###### ${content.trim()}\n\n`;
+        case "p": return ctx.inTable ? content : `${content.trim()}\n\n`;
+        case "code": {
+          const codeContent = children ? extractNode(children as Node, { ...ctx, inCode: true }) : "";
+          return ctx.inCode ? codeContent : `\`${codeContent}\``;
+        }
+        case "pre": {
+          const preContent = children ? extractNode(children as Node, { ...ctx, inCode: true }) : "";
+          return "```\n" + preContent + "\n```\n\n";
+        }
+        case "strong": case "b": return `**${content}**`;
+        case "em": case "i": return `*${content}*`;
+        case "li": return `- ${content.trim()}\n`;
+        case "ul": case "ol": return content + "\n";
+        case "blockquote": return `> ${content.trim()}\n\n`;
+        case "table": return extractTable(node as unknown[]) + "\n";
+        case "thead": case "tbody": case "tr": case "th": case "td":
+          return content;
+        case "div":
+          if (props.role === "alert" || props["data-slot"] === "alert") {
+            return `> ${content.trim()}\n\n`;
+          }
+          return content;
+        case "a": {
+          const href = props.href as string | undefined;
+          return href && !href.startsWith("#") ? `[${content}](${href})` : content;
+        }
+        default: return content;
+      }
+    }
+
+    // Array of child nodes
+    return (node as Node[]).map(n => extractNode(n, ctx)).join("");
+  }
+
+  function extractTable(tableNode: unknown[]): string {
+    const props = (tableNode[3] || {}) as Record<string, unknown>;
+    const rows: string[][] = [];
+    let headerRowCount = 0;
+
+    function walkTable(node: unknown, isHeader = false): void {
+      if (node === null || node === undefined) return;
+      
+      // Handle string refs
+      if (typeof node === "string") {
+        const refMatch = node.match(/^\$L([0-9a-f]+)$/i);
+        if (refMatch && !visitedRefs.has(refMatch[1])) {
+          visitedRefs.add(refMatch[1]);
+          const refNode = getParsedChunk(refMatch[1]);
+          if (refNode) walkTable(refNode, isHeader);
+          visitedRefs.delete(refMatch[1]);
+        }
+        return;
+      }
+      
+      if (!Array.isArray(node)) return;
+      
+      if (node[0] === "$") {
+        const tag = node[1] as string;
+        const nodeProps = (node[3] || {}) as Record<string, unknown>;
+        
+        // Handle component refs
+        if (tag.startsWith("$L")) {
+          const refId = tag.slice(2);
+          if (!visitedRefs.has(refId)) {
+            visitedRefs.add(refId);
+            const refNode = getParsedChunk(refId);
+            if (refNode) walkTable(refNode, isHeader);
+            visitedRefs.delete(refId);
+          }
+          return;
+        }
+        
+        if (tag === "thead") walkTable(nodeProps.children, true);
+        else if (tag === "tbody") walkTable(nodeProps.children, false);
+        else if (tag === "tr") {
+          const cells: string[] = [];
+          walkCells(nodeProps.children, cells);
+          if (cells.length > 0) {
+            rows.push(cells);
+            if (isHeader) headerRowCount++;
+          }
+        } else walkTable(nodeProps.children, isHeader);
+      } else {
+        for (const child of node) walkTable(child, isHeader);
+      }
+    }
+
+    function walkCells(node: unknown, cells: string[]): void {
+      if (node === null || node === undefined) return;
+      
+      // Handle string refs
+      if (typeof node === "string") {
+        const refMatch = node.match(/^\$L([0-9a-f]+)$/i);
+        if (refMatch && !visitedRefs.has(refMatch[1])) {
+          visitedRefs.add(refMatch[1]);
+          const refNode = getParsedChunk(refMatch[1]);
+          if (refNode) walkCells(refNode, cells);
+          visitedRefs.delete(refMatch[1]);
+        }
+        return;
+      }
+      
+      if (!Array.isArray(node)) return;
+      
+      if (node[0] === "$" && (node[1] === "td" || node[1] === "th")) {
+        const cellProps = (node[3] || {}) as Record<string, unknown>;
+        const text = extractNode(cellProps.children, { inTable: true, inCode: false })
+          .trim()
+          .replace(/\n/g, " ")
+          .replace(/\\/g, "\\\\")  // Escape backslashes first
+          .replace(/\|/g, "\\|");  // Then escape pipes
+        cells.push(text);
+      } else if (node[0] === "$" && typeof node[1] === "string" && (node[1] as string).startsWith("$L")) {
+        // Component ref for a cell
+        const refId = (node[1] as string).slice(2);
+        if (!visitedRefs.has(refId)) {
+          visitedRefs.add(refId);
+          const refNode = getParsedChunk(refId);
+          if (refNode) walkCells(refNode, cells);
+          visitedRefs.delete(refId);
+        }
+      } else {
+        for (const child of node) walkCells(child, cells);
+      }
+    }
+
+    walkTable(props.children);
+    if (rows.length === 0) return "";
+
+    const colCount = Math.max(...rows.map(r => r.length));
+    let md = "";
+    for (let i = 0; i < rows.length; i++) {
+      const row = rows[i].concat(Array(colCount - rows[i].length).fill(""));
+      md += "| " + row.join(" | ") + " |\n";
+      if (i === headerRowCount - 1 || (headerRowCount === 0 && i === 0)) {
+        md += "| " + Array(colCount).fill("---").join(" | ") + " |\n";
+      }
+    }
+    return md;
+  }
+
+  // Process main content chunk (usually "23")
+  const mainChunk = getParsedChunk("23");
+  
+  if (mainChunk) {
+    const content = extractNode(mainChunk);
+    if (content.trim().length > 100) {
+      const cleaned = content
+        .replace(/\n{3,}/g, "\n\n")
+        .trim();
+      return { title, content: cleaned };
+    }
+  }
+
+  // Fallback: try other chunks
+  const contentParts: { order: number; text: string }[] = [];
+
+  for (const [id] of chunkMap) {
+    if (id === "23") continue;
+    const parsed = getParsedChunk(id);
+    if (!parsed) continue;
+
+    visitedRefs.clear();
+    const text = extractNode(parsed);
+
+    if (text.trim().length > 50 && 
+        !text.includes("page was not found") && 
+        !text.includes("404")) {
+      contentParts.push({ order: parseInt(id, 16), text: text.trim() });
+    }
+  }
+
+  if (contentParts.length === 0) return null;
+
+  contentParts.sort((a, b) => a.order - b.order);
+  
+  const seen = new Set<string>();
+  const uniqueParts: string[] = [];
+  for (const part of contentParts) {
+    const key = part.text.slice(0, 150);
+    if (!seen.has(key)) {
+      seen.add(key);
+      uniqueParts.push(part.text);
+    }
+  }
+
+  const content = uniqueParts.join("\n\n").replace(/\n{3,}/g, "\n\n").trim();
+  return content.length > 100 ? { title, content } : null;
+}

package/search.ts

@@ -0,0 +1,49 @@
+import { activityMonitor } from "./activity.js";
+import { hasExaApiKey, searchWithExa } from "./exa.js";
+import type { SearchResponse, SearchOptions } from "./types.js";
+
+const MAX_NUM_RESULTS = 20;
+
+export interface FullSearchOptions extends SearchOptions {
+	includeContent?: boolean;
+}
+
+function errorMessage(err: unknown): string {
+	return err instanceof Error ? err.message : String(err);
+}
+
+function isAbortError(err: unknown): boolean {
+	return errorMessage(err).toLowerCase().includes("abort");
+}
+
+export async function search(query: string, options: FullSearchOptions = {}): Promise<SearchResponse> {
+	const activityId = activityMonitor.logStart({ type: "api", query });
+
+	try {
+		const result = await searchWithExa(query, options);
+		if (result && "exhausted" in result) {
+			throw new Error(
+				"Exa monthly free tier exhausted (1,000 requests). Resets next month.\n" +
+				"  Upgrade at exa.ai/pricing"
+			);
+		}
+		if (result && "answer" in result) {
+			activityMonitor.logComplete(activityId, 200);
+			return result;
+		}
+		// null result from MCP with no API key
+		throw new Error(
+			"No search provider available. Either:\n" +
+			"  1. Set EXA_API_KEY (or exaApiKey in ~/.pi/web-search.json)\n" +
+			"  2. Use Exa MCP (no API key needed)"
+		);
+	} catch (err) {
+		const message = err instanceof Error ? err.message : String(err);
+		if (isAbortError(err)) {
+			activityMonitor.logComplete(activityId, 0);
+		} else {
+			activityMonitor.logError(activityId, message);
+		}
+		throw err;
+	}
+}

package/storage.ts

@@ -0,0 +1,71 @@
+import type { ExtensionContext } from "@mariozechner/pi-coding-agent";
+import type { ExtractedContent } from "./extract.js";
+import type { SearchResult } from "./types.js";
+
+const CACHE_TTL_MS = 60 * 60 * 1000;
+
+export interface QueryResultData {
+	query: string;
+	answer: string;
+	results: SearchResult[];
+	error: string | null;
+}
+
+export interface StoredSearchData {
+	id: string;
+	type: "search" | "fetch";
+	timestamp: number;
+	queries?: QueryResultData[];
+	urls?: ExtractedContent[];
+}
+
+const storedResults = new Map<string, StoredSearchData>();
+
+export function generateId(): string {
+	return Date.now().toString(36) + Math.random().toString(36).slice(2, 8);
+}
+
+export function storeResult(id: string, data: StoredSearchData): void {
+	storedResults.set(id, data);
+}
+
+export function getResult(id: string): StoredSearchData | null {
+	return storedResults.get(id) ?? null;
+}
+
+export function getAllResults(): StoredSearchData[] {
+	return Array.from(storedResults.values());
+}
+
+export function deleteResult(id: string): boolean {
+	return storedResults.delete(id);
+}
+
+export function clearResults(): void {
+	storedResults.clear();
+}
+
+function isValidStoredData(data: unknown): data is StoredSearchData {
+	if (!data || typeof data !== "object") return false;
+	const d = data as Record<string, unknown>;
+	if (typeof d.id !== "string" || !d.id) return false;
+	if (d.type !== "search" && d.type !== "fetch") return false;
+	if (typeof d.timestamp !== "number") return false;
+	if (d.type === "search" && !Array.isArray(d.queries)) return false;
+	if (d.type === "fetch" && !Array.isArray(d.urls)) return false;
+	return true;
+}
+
+export function restoreFromSession(ctx: ExtensionContext): void {
+	storedResults.clear();
+	const now = Date.now();
+
+	for (const entry of ctx.sessionManager.getBranch()) {
+		if (entry.type === "custom" && entry.customType === "web-search-results") {
+			const data = entry.data;
+			if (isValidStoredData(data) && now - data.timestamp < CACHE_TTL_MS) {
+				storedResults.set(data.id, data);
+			}
+		}
+	}
+}