Export Perplexity.ai conversations as markdown with configurable citation styles
目前為
// ==UserScript==
// @name Perplexity.ai Chat Exporter
// @namespace https://github.com/ckep1/pplxport
// @version 2.0.0
// @description Export Perplexity.ai conversations as markdown with configurable citation styles
// @author Chris Kephart
// @match https://www.perplexity.ai/*
// @grant GM_getValue
// @grant GM_setValue
// @grant GM_registerMenuCommand
// @run-at document-idle
// @license MIT
// ==/UserScript==
(function () {
"use strict";
// ============================================================================
// CONFIGURATION & CONSTANTS
// ============================================================================
const DEBUG = false;
const console = DEBUG ? window.console : { log() {}, warn() {}, error() {} };
// Style options
const CITATION_STYLES = {
ENDNOTES: "endnotes",
INLINE: "inline",
PARENTHESIZED: "parenthesized",
NAMED: "named",
};
const FORMAT_STYLES = {
FULL: "full", // Include User/Assistant tags and all dividers
CONCISE: "concise", // Just content, minimal dividers
};
// Global citation tracking for consistent numbering across all responses
const globalCitations = {
urlToNumber: new Map(), // normalized URL -> citation number
citationRefs: new Map(), // citation number -> {href, sourceName, normalizedUrl}
nextCitationNumber: 1,
reset() {
this.urlToNumber.clear();
this.citationRefs.clear();
this.nextCitationNumber = 1;
},
addCitation(url, sourceName = null) {
const normalizedUrl = normalizeUrl(url);
if (!this.urlToNumber.has(normalizedUrl)) {
this.urlToNumber.set(normalizedUrl, this.nextCitationNumber);
this.citationRefs.set(this.nextCitationNumber, {
href: url,
sourceName,
normalizedUrl,
});
this.nextCitationNumber++;
}
return this.urlToNumber.get(normalizedUrl);
},
getCitationNumber(url) {
const normalizedUrl = normalizeUrl(url);
return this.urlToNumber.get(normalizedUrl);
},
};
// ============================================================================
// MENU COMMANDS
// ============================================================================
GM_registerMenuCommand("Use Endnotes Citation Style", () => {
GM_setValue("citationStyle", CITATION_STYLES.ENDNOTES);
alert("Citation style set to endnotes. Format: [1] with sources listed at end.");
});
GM_registerMenuCommand("Use Inline Citation Style", () => {
GM_setValue("citationStyle", CITATION_STYLES.INLINE);
alert("Citation style set to inline. Format: [1](url)");
});
GM_registerMenuCommand("Use Parenthesized Citation Style", () => {
GM_setValue("citationStyle", CITATION_STYLES.PARENTHESIZED);
alert("Citation style set to parenthesized. Format: ([1](url))");
});
GM_registerMenuCommand("Use Named Citation Style", () => {
GM_setValue("citationStyle", CITATION_STYLES.NAMED);
alert("Citation style set to named. Format: [source](url) using source names like 'wikipedia', 'reddit', etc.");
});
GM_registerMenuCommand("Full Format (with User/Assistant)", () => {
GM_setValue("formatStyle", FORMAT_STYLES.FULL);
alert("Format set to full with User/Assistant tags.");
});
GM_registerMenuCommand("Concise Format (content only)", () => {
GM_setValue("formatStyle", FORMAT_STYLES.CONCISE);
alert("Format set to concise content only.");
});
GM_registerMenuCommand("Enable Extra Newlines", () => {
GM_setValue("addExtraNewlines", true);
alert("Extra newlines enabled. Adds blank lines after paragraphs and list items.");
});
GM_registerMenuCommand("Disable Extra Newlines", () => {
GM_setValue("addExtraNewlines", false);
alert("Extra newlines disabled. Standard markdown spacing.");
});
// ============================================================================
// UTILITY FUNCTIONS
// ============================================================================
// Get user preferences
function getPreferences() {
return {
citationStyle: GM_getValue("citationStyle", CITATION_STYLES.PARENTHESIZED),
formatStyle: GM_getValue("formatStyle", FORMAT_STYLES.FULL),
addExtraNewlines: GM_getValue("addExtraNewlines", false),
};
}
// Extract source name from text, handling various formats
function extractSourceName(text) {
if (!text) return null;
// Clean the text
text = text.trim();
// If it's a pattern like "rabbit+2", "reddit+1", extract the source name
const plusMatch = text.match(/^([a-zA-Z]+)\+\d+$/);
if (plusMatch) {
return plusMatch[1];
}
// If it's just text without numbers, use it as is (but clean it up)
const cleanName = text.replace(/[^a-zA-Z0-9-_]/g, "").toLowerCase();
if (cleanName && cleanName.length > 0) {
return cleanName;
}
return null;
}
// Normalize URL by removing fragments (#) to group same page citations
function normalizeUrl(url) {
if (!url) return null;
try {
const urlObj = new URL(url);
// Remove the fragment (hash) portion
urlObj.hash = "";
return urlObj.toString();
} catch (e) {
// If URL parsing fails, just remove # manually
return url.split("#")[0];
}
}
// Extract domain name from URL for named citations
function extractDomainName(url) {
if (!url) return null;
try {
const urlObj = new URL(url);
let domain = urlObj.hostname.toLowerCase();
// Remove www. prefix
domain = domain.replace(/^www\./, "");
// Get the main domain part (before first dot for common cases)
const parts = domain.split(".");
if (parts.length >= 2) {
// Handle special cases like co.uk, github.io, etc.
if (parts[parts.length - 2].length <= 3 && parts.length > 2) {
return parts[parts.length - 3];
} else {
return parts[parts.length - 2];
}
}
return parts[0];
} catch (e) {
return null;
}
}
// ============================================================================
// DOM HELPER FUNCTIONS
// ============================================================================
function getThreadContainer() {
return document.querySelector('.max-w-threadContentWidth, [class*="threadContentWidth"]') || document.querySelector("main") || document.body;
}
function getScrollRoot() {
const thread = getThreadContainer();
const candidates = [];
let node = thread;
while (node && node !== document.body) {
candidates.push(node);
node = node.parentElement;
}
const scrollingElement = document.scrollingElement || document.documentElement;
candidates.push(scrollingElement);
let best = null;
for (const el of candidates) {
try {
const style = getComputedStyle(el);
const overflowY = (style.overflowY || style.overflow || "").toLowerCase();
const canScroll = el.scrollHeight - el.clientHeight > 50;
const isScrollable = /auto|scroll|overlay/.test(overflowY) || el === scrollingElement;
if (canScroll && isScrollable) {
if (!best || el.scrollHeight > best.scrollHeight) {
best = el;
}
}
} catch (e) {
// ignore
}
}
return best || scrollingElement;
}
function isInViewport(el, margin = 8) {
const rect = el.getBoundingClientRect();
const vh = window.innerHeight || document.documentElement.clientHeight;
const vw = window.innerWidth || document.documentElement.clientWidth;
return rect.bottom > -margin && rect.top < vh + margin && rect.right > -margin && rect.left < vw + margin;
}
function isCodeCopyButton(btn) {
const testId = btn.getAttribute("data-testid");
const ariaLower = (btn.getAttribute("aria-label") || "").toLowerCase();
if (testId === "copy-code-button" || testId === "copy-code" || (testId && testId.includes("copy-code"))) return true;
if (ariaLower.includes("copy code")) return true;
if (btn.closest("pre") || btn.closest("code")) return true;
return false;
}
function findUserMessageRootFromElement(el) {
let node = el;
let depth = 0;
while (node && node !== document.body && depth < 10) {
if (node.querySelector && (node.querySelector("button[data-testid='copy-query-button']") || node.querySelector("button[aria-label='Copy Query']") || node.querySelector("span[data-lexical-text='true']"))) {
return node;
}
node = node.parentElement;
depth++;
}
return el.parentElement || el;
}
function findUserMessageRootFrom(button) {
let node = button;
let depth = 0;
while (node && node !== document.body && depth < 10) {
// A user message root should contain lexical text from the input/query
if (node.querySelector && (node.querySelector(".whitespace-pre-line.text-pretty.break-words") || node.querySelector("span[data-lexical-text='true']"))) {
return node;
}
node = node.parentElement;
depth++;
}
return button.parentElement || button;
}
function findAssistantMessageRootFrom(button) {
let node = button;
let depth = 0;
while (node && node !== document.body && depth < 10) {
// An assistant message root should contain the prose answer block
if (node.querySelector && node.querySelector(".prose.text-pretty.dark\\:prose-invert, [class*='prose'][class*='prose-invert'], [data-testid='answer'], [data-testid='assistant']")) {
return node;
}
node = node.parentElement;
depth++;
}
return button.parentElement || button;
}
// ============================================================================
// SCROLL & NAVIGATION HELPERS
// ============================================================================
async function pageDownOnce(scroller, delayMs = 500, factor = 0.9) {
if (!scroller) scroller = getScrollRoot();
const delta = Math.max(200, Math.floor(scroller.clientHeight * factor));
scroller.scrollTop = Math.min(scroller.scrollTop + delta, scroller.scrollHeight);
await new Promise((r) => setTimeout(r, delayMs));
}
async function preloadPageFully() {
try {
const scroller = getScrollRoot();
window.focus();
scroller.scrollTop = 0;
await new Promise((resolve) => setTimeout(resolve, 300));
let lastHeight = scroller.scrollHeight;
let stableCount = 0;
const maxTries = 80; // ~40s at 500ms intervals
for (let i = 0; i < maxTries && stableCount < 3; i++) {
scroller.scrollTop = scroller.scrollHeight;
await new Promise((resolve) => setTimeout(resolve, 500));
const newHeight = scroller.scrollHeight;
if (newHeight > lastHeight + 10) {
lastHeight = newHeight;
stableCount = 0;
} else {
stableCount++;
}
}
// Return to top so processing starts from the beginning
scroller.scrollTop = 0;
await new Promise((resolve) => setTimeout(resolve, 500));
} catch (e) {
// Non-fatal; we'll just proceed
console.warn("Preload scroll encountered an issue:", e);
}
}
function simulateHover(element) {
try {
const rect = element.getBoundingClientRect();
const x = rect.left + Math.min(20, Math.max(2, rect.width / 3));
const y = rect.top + Math.min(20, Math.max(2, rect.height / 3));
const opts = { bubbles: true, clientX: x, clientY: y };
element.dispatchEvent(new MouseEvent("mouseenter", opts));
element.dispatchEvent(new MouseEvent("mouseover", opts));
element.dispatchEvent(new MouseEvent("mousemove", opts));
} catch (e) {
// best effort
}
}
async function readClipboardWithRetries(maxRetries = 3, delayMs = 200) {
let last = "";
for (let i = 0; i < maxRetries; i++) {
try {
const text = await navigator.clipboard.readText();
if (text && text.trim() && text !== last) {
return text;
}
last = text;
} catch (e) {
// keep retrying
}
await new Promise((r) => setTimeout(r, delayMs));
}
try {
return await navigator.clipboard.readText();
} catch {
return "";
}
}
// Click expanders like "Show more", "Read more", etc. Best-effort
const clickedExpanders = new WeakSet();
function findExpanders(limit = 8) {
const candidates = [];
const patterns = /(show more|read more|view more|see more|expand|load more|view full|show all|continue reading)/i;
const els = document.querySelectorAll('button, a, [role="button"]');
for (const el of els) {
if (candidates.length >= limit) break;
if (clickedExpanders.has(el)) continue;
const label = (el.getAttribute("aria-label") || "").trim();
const text = (el.textContent || "").trim();
if (patterns.test(label) || patterns.test(text)) {
// avoid code-block related buttons
if (el.closest("pre, code")) continue;
candidates.push(el);
}
}
return candidates;
}
async function clickExpandersOnce(limit = 6) {
const expanders = findExpanders(limit);
if (expanders.length === 0) return false;
for (const el of expanders) {
try {
clickedExpanders.add(el);
el.dispatchEvent(new MouseEvent("mouseenter", { bubbles: true }));
await new Promise((r) => setTimeout(r, 50));
el.click();
await new Promise((r) => setTimeout(r, 150));
} catch {}
}
// allow expanded content to render
await new Promise((r) => setTimeout(r, 250));
return true;
}
// ============================================================================
// BUTTON HELPER FUNCTIONS
// ============================================================================
function getViewportQueryButtons() {
const buttons = Array.from(document.querySelectorAll('button[data-testid="copy-query-button"], button[aria-label="Copy Query"]'));
return buttons.filter((btn) => isInViewport(btn) && !btn.closest("pre,code"));
}
function getViewportResponseButtons() {
const buttons = Array.from(document.querySelectorAll('button[aria-label="Copy"]')).filter((btn) => btn.querySelector("svg.tabler-icon-copy"));
return buttons.filter((btn) => isInViewport(btn) && !btn.closest("pre,code"));
}
async function clickVisibleButtonAndGetClipboard(button) {
try {
window.focus();
simulateHover(button);
await new Promise((r) => setTimeout(r, 100));
button.focus();
button.click();
await new Promise((r) => setTimeout(r, 200));
return await readClipboardWithRetries(3, 150);
} catch (e) {
return "";
}
}
async function clickButtonAndGetClipboard(button) {
window.focus();
button.scrollIntoView({ behavior: "smooth", block: "center", inline: "center" });
await new Promise((r) => setTimeout(r, 200));
simulateHover(button);
await new Promise((r) => setTimeout(r, 150));
button.focus();
button.click();
await new Promise((r) => setTimeout(r, 500));
window.focus();
return await readClipboardWithRetries(4, 250);
}
function collectAnchoredMessageRootsOnce() {
const roots = new Map(); // rootEl -> { rootEl, top, queryButton, responseButton }
const queryButtons = Array.from(document.querySelectorAll('button[data-testid="copy-query-button"], button[aria-label="Copy Query"]'));
for (const btn of queryButtons) {
if (isCodeCopyButton(btn)) continue;
const root = findUserMessageRootFrom(btn);
const top = root.getBoundingClientRect().top + window.scrollY || btn.getBoundingClientRect().top + window.scrollY;
const obj = roots.get(root) || { rootEl: root, top, queryButton: null, responseButton: null };
obj.queryButton = obj.queryButton || btn;
obj.top = Math.min(obj.top, top);
roots.set(root, obj);
}
const responseButtons = Array.from(document.querySelectorAll('button[aria-label="Copy"]')).filter((btn) => btn.querySelector("svg.tabler-icon-copy"));
for (const btn of responseButtons) {
if (isCodeCopyButton(btn)) continue;
const root = findAssistantMessageRootFrom(btn);
// Ensure the root actually holds an assistant answer, not some header copy control
const hasAnswer = !!root.querySelector(".prose.text-pretty.dark\\:prose-invert, [class*='prose'][class*='prose-invert']");
if (!hasAnswer) continue;
const top = root.getBoundingClientRect().top + window.scrollY || btn.getBoundingClientRect().top + window.scrollY;
const obj = roots.get(root) || { rootEl: root, top, queryButton: null, responseButton: null };
obj.responseButton = obj.responseButton || btn;
obj.top = Math.min(obj.top, top);
roots.set(root, obj);
}
return Array.from(roots.values()).sort((a, b) => a.top - b.top);
}
// ============================================================================
// EXTRACTION METHODS - ALL GROUPED TOGETHER
// ============================================================================
// Method 1: Page-down with button clicking (most reliable)
async function extractByPageDownClickButtons(citationStyle) {
const conversation = [];
const processedContent = new Set();
const processedQueryButtons = new WeakSet();
const processedAnswerButtons = new WeakSet();
const scroller = getScrollRoot();
scroller.scrollTop = 0;
await new Promise((r) => setTimeout(r, 200));
let stableBottomCount = 0;
let scrollAttempt = 0;
const maxScrollAttempts = 300;
const scrollDelay = 500;
while (scrollAttempt < maxScrollAttempts && stableBottomCount < 5) {
scrollAttempt++;
let processedSomething = false;
// Collect visible query/response copy buttons and process in top-to-bottom order
const qButtons = getViewportQueryButtons().map((btn) => ({ btn, role: "User" }));
const rButtons = getViewportResponseButtons().map((btn) => ({ btn, role: "Assistant" }));
const allButtons = [...qButtons, ...rButtons].sort((a, b) => {
const at = a.btn.getBoundingClientRect().top;
const bt = b.btn.getBoundingClientRect().top;
return at - bt;
});
for (const item of allButtons) {
const { btn, role } = item;
if (role === "User") {
if (processedQueryButtons.has(btn)) continue;
processedQueryButtons.add(btn);
const text = (await clickVisibleButtonAndGetClipboard(btn))?.trim();
if (text) {
const hash = text.substring(0, 200) + text.substring(Math.max(0, text.length - 50)) + text.length + "|U";
if (!processedContent.has(hash)) {
processedContent.add(hash);
conversation.push({ role: "User", content: text });
processedSomething = true;
}
}
} else {
if (processedAnswerButtons.has(btn)) continue;
processedAnswerButtons.add(btn);
const raw = (await clickVisibleButtonAndGetClipboard(btn))?.trim();
if (raw) {
const hash = raw.substring(0, 200) + raw.substring(Math.max(0, raw.length - 50)) + raw.length;
if (!processedContent.has(hash)) {
processedContent.add(hash);
const processedMarkdown = processCopiedMarkdown(raw, citationStyle);
conversation.push({ role: "Assistant", content: processedMarkdown });
processedSomething = true;
}
}
}
}
// Expand any collapsed content every few steps if nothing was processed
if (!processedSomething) {
await clickExpandersOnce(6);
}
const beforeBottom = scroller.scrollTop + scroller.clientHeight >= scroller.scrollHeight - 2;
await pageDownOnce(scroller, scrollDelay, 0.9);
const afterBottom = scroller.scrollTop + scroller.clientHeight >= scroller.scrollHeight - 2;
if (beforeBottom && afterBottom && !processedSomething) {
stableBottomCount++;
} else {
stableBottomCount = 0;
}
}
return conversation;
}
// Method 2: Single-pass DOM scan (no button clicking)
async function extractByDomScanSinglePass(citationStyle) {
const processedContent = new Set();
const collected = [];
const scroller = getScrollRoot();
scroller.scrollTop = 0;
await new Promise((r) => setTimeout(r, 300));
let stableBottomCount = 0;
let scrollAttempt = 0;
const maxScrollAttempts = 300;
const scrollDelay = 500;
while (scrollAttempt < maxScrollAttempts && stableBottomCount < 5) {
scrollAttempt++;
const beforeCount = collected.length;
// Collect in DOM order for this viewport/state
const batch = collectDomMessagesInOrderOnce(citationStyle, processedContent);
if (batch.length > 0) {
for (const item of batch) {
collected.push(item);
}
} else {
// Try expanding collapsed sections and collect again
const expanded = await clickExpandersOnce(8);
if (expanded) {
const batch2 = collectDomMessagesInOrderOnce(citationStyle, processedContent);
if (batch2.length > 0) {
for (const item of batch2) collected.push(item);
}
}
}
// Detect bottom
const atBottom = scroller.scrollTop + scroller.clientHeight >= scroller.scrollHeight - 2;
await pageDownOnce(scroller, scrollDelay, 0.9);
const atBottomAfter = scroller.scrollTop + scroller.clientHeight >= scroller.scrollHeight - 2;
if (atBottom && atBottomAfter && collected.length === beforeCount) {
stableBottomCount++;
} else {
stableBottomCount = 0;
}
}
// Do not return to top; keep scroller where it ended
return collected;
}
// Helper for Method 2: collect messages in DOM order within a pass
function collectDomMessagesInOrderOnce(citationStyle, processedContent) {
const results = [];
const container = getThreadContainer();
const assistantSelector = ".prose.text-pretty.dark\\:prose-invert, [class*='prose'][class*='prose-invert']";
const userSelectors = [".whitespace-pre-line.text-pretty.break-words", ".group\\/query span[data-lexical-text='true']", "h1.group\\/query span[data-lexical-text='true']", "span[data-lexical-text='true']"];
const combined = `${assistantSelector}, ${userSelectors.join(", ")}`;
const nodes = container.querySelectorAll(combined);
nodes.forEach((node) => {
if (node.matches(assistantSelector)) {
const cloned = node.cloneNode(true);
const md = htmlToMarkdown(cloned.innerHTML, citationStyle).trim();
if (!md) return;
const hash = md.substring(0, 200) + md.substring(Math.max(0, md.length - 50)) + md.length;
if (processedContent.has(hash)) return;
processedContent.add(hash);
results.push({ role: "Assistant", content: md });
} else {
// User
const root = findUserMessageRootFromElement(node);
if (root.closest && (root.closest(".prose.text-pretty.dark\\:prose-invert") || root.closest("[class*='prose'][class*='prose-invert']"))) return;
// Aggregate query text from all lexical spans within the same root for stability
const spans = root.querySelectorAll("span[data-lexical-text='true']");
let text = "";
if (spans.length > 0) {
text = Array.from(spans)
.map((s) => (s.textContent || "").trim())
.join(" ")
.trim();
} else {
text = (node.textContent || "").trim();
}
if (!text || text.length < 2) return;
// Prefer nodes within a container that also has a copy-query button, but don't require it
const hasCopyQueryButton = !!(root.querySelector && (root.querySelector("button[data-testid='copy-query-button']") || root.querySelector("button[aria-label='Copy Query']")));
if (!hasCopyQueryButton && text.length < 10) return;
const hash = text.substring(0, 200) + text.substring(Math.max(0, text.length - 50)) + text.length + "|U";
if (processedContent.has(hash)) return;
processedContent.add(hash);
results.push({ role: "User", content: text });
}
});
return results;
}
// Method 3: Anchored copy button approach (more complex, uses scrollIntoView)
async function extractUsingCopyButtons(citationStyle) {
// Reset global citation tracking for this export
globalCitations.reset();
try {
// First try anchored, container-aware approach with preload + progressive scroll
const anchored = await processAnchoredButtonsWithProgressiveScroll(citationStyle);
if (anchored.length > 0) {
return anchored;
}
// Fallback: robust scroll-and-process (legacy)
return await scrollAndProcessButtons(citationStyle);
} catch (e) {
console.error("Copy button extraction failed:", e);
return [];
}
}
async function processAnchoredButtonsWithProgressiveScroll(citationStyle) {
const conversation = [];
const processedContent = new Set();
const processedButtons = new WeakSet();
await preloadPageFully();
// Start at top and progressively page down to handle virtualized lists
const scroller = getScrollRoot();
scroller.scrollTop = 0;
await new Promise((r) => setTimeout(r, 400));
let stableCount = 0;
let scrollAttempt = 0;
const maxScrollAttempts = 120;
const scrollDelay = 600;
while (scrollAttempt < maxScrollAttempts && stableCount < 5) {
scrollAttempt++;
const roots = collectAnchoredMessageRootsOnce();
let processedSomethingThisPass = false;
for (const item of roots) {
const { queryButton, responseButton } = item;
// Process query first
if (queryButton && !processedButtons.has(queryButton)) {
try {
const text = (await clickButtonAndGetClipboard(queryButton))?.trim();
if (text) {
const contentHash = text.substring(0, 200) + text.substring(Math.max(0, text.length - 50)) + text.length;
if (!processedContent.has(contentHash)) {
processedContent.add(contentHash);
conversation.push({ role: "User", content: text });
processedSomethingThisPass = true;
}
}
} catch (e) {
console.warn("Query copy failed:", e);
} finally {
processedButtons.add(queryButton);
}
}
// Then process response
if (responseButton && !processedButtons.has(responseButton)) {
try {
const raw = (await clickButtonAndGetClipboard(responseButton))?.trim();
if (raw) {
const contentHash = raw.substring(0, 200) + raw.substring(Math.max(0, raw.length - 50)) + raw.length;
if (!processedContent.has(contentHash)) {
processedContent.add(contentHash);
const processedMarkdown = processCopiedMarkdown(raw, citationStyle);
conversation.push({ role: "Assistant", content: processedMarkdown });
processedSomethingThisPass = true;
}
}
} catch (e) {
console.warn("Response copy failed:", e);
} finally {
processedButtons.add(responseButton);
}
}
}
if (!processedSomethingThisPass) {
stableCount++;
} else {
stableCount = 0;
}
// Page down and allow DOM to settle
await pageDownOnce(scroller, scrollDelay, 0.9);
}
// Try to catch any remaining at the end with a final full scan without scrolling
const finalRoots = collectAnchoredMessageRootsOnce();
for (const { queryButton, responseButton } of finalRoots) {
if (queryButton && !processedButtons.has(queryButton)) {
try {
const text = (await clickButtonAndGetClipboard(queryButton))?.trim();
if (text) {
const contentHash = text.substring(0, 200) + text.substring(Math.max(0, text.length - 50)) + text.length;
if (!processedContent.has(contentHash)) {
processedContent.add(contentHash);
conversation.push({ role: "User", content: text });
}
}
} catch {}
}
if (responseButton && !processedButtons.has(responseButton)) {
try {
const raw = (await clickButtonAndGetClipboard(responseButton))?.trim();
if (raw) {
const contentHash = raw.substring(0, 200) + raw.substring(Math.max(0, raw.length - 50)) + raw.length;
if (!processedContent.has(contentHash)) {
processedContent.add(contentHash);
const processedMarkdown = processCopiedMarkdown(raw, citationStyle);
conversation.push({ role: "Assistant", content: processedMarkdown });
}
}
} catch {}
}
}
// Return to top
scroller.scrollTop = 0;
await new Promise((r) => setTimeout(r, 300));
return conversation;
}
// Robustly scroll through page and process copy buttons as we find them
async function scrollAndProcessButtons(citationStyle) {
console.log("Starting robust scroll and process...");
const conversation = [];
const processedContent = new Set();
const processedButtons = new Set();
// Ensure document stays focused
window.focus();
// Start from top
const scroller = getScrollRoot();
scroller.scrollTop = 0;
await new Promise((resolve) => setTimeout(resolve, 500));
let stableCount = 0;
let scrollAttempt = 0;
let lastButtonCount = 0;
const maxScrollAttempts = 100; // Increase max attempts
const scrollDelay = 600; // delay between page downs
while (scrollAttempt < maxScrollAttempts && stableCount < 5) {
scrollAttempt++;
// Count current buttons before processing
const currentButtonCount = document.querySelectorAll('button[data-testid="copy-query-button"], button[aria-label="Copy Query"], button[aria-label="Copy"]').length;
console.log(`Page Down attempt ${scrollAttempt}: buttons=${currentButtonCount}`);
// Find and process visible copy buttons at current position
await processVisibleButtons();
// Track button count changes
if (currentButtonCount > lastButtonCount) {
console.log(`Button count increased from ${lastButtonCount} to ${currentButtonCount}`);
lastButtonCount = currentButtonCount;
stableCount = 0; // Reset stability when new buttons found
} else {
stableCount++;
console.log(`Button count stable at ${currentButtonCount} (stability: ${stableCount}/5)`);
}
// Page down the actual scroller
await pageDownOnce(scroller, scrollDelay, 0.9);
}
console.log(`Scroll complete after ${scrollAttempt} attempts. Found ${conversation.length} conversation items`);
// Return to top
scroller.scrollTop = 0;
await new Promise((resolve) => setTimeout(resolve, 500));
return conversation;
// Helper function to process visible buttons at current scroll position
async function processVisibleButtons() {
const allButtons = document.querySelectorAll("button");
const copyButtons = [];
allButtons.forEach((btn) => {
if (processedButtons.has(btn)) return;
// Exclude code block copy buttons
const testId = btn.getAttribute("data-testid");
const ariaLower = (btn.getAttribute("aria-label") || "").toLowerCase();
if (testId === "copy-code-button" || testId === "copy-code" || testId?.includes("copy-code") || ariaLower.includes("copy code") || btn.closest("pre") || btn.closest("code")) {
return;
}
// Only include conversation copy buttons
const isQueryCopyButton = testId === "copy-query-button" || btn.getAttribute("aria-label") === "Copy Query";
const isResponseCopyButton = btn.getAttribute("aria-label") === "Copy" && btn.querySelector("svg.tabler-icon-copy");
if (isQueryCopyButton) {
copyButtons.push({ el: btn, role: "User" });
} else if (isResponseCopyButton) {
copyButtons.push({ el: btn, role: "Assistant" });
}
});
// Sort by vertical position
copyButtons.sort((a, b) => {
const aTop = a.el.getBoundingClientRect().top + window.scrollY;
const bTop = b.el.getBoundingClientRect().top + window.scrollY;
return aTop - bTop;
});
console.log(`Found ${copyButtons.length} copy buttons in DOM`);
// Process each copy button (don't filter by viewport visibility)
for (const { el: button, role } of copyButtons) {
if (processedButtons.has(button)) continue;
try {
processedButtons.add(button);
// Ensure window stays focused
window.focus();
// Scroll button into view and center it
button.scrollIntoView({
behavior: "smooth",
block: "center",
inline: "center",
});
await new Promise((resolve) => setTimeout(resolve, 300));
// Click button
button.focus();
button.click();
// Wait for clipboard
await new Promise((resolve) => setTimeout(resolve, 500));
window.focus();
const clipboardText = await navigator.clipboard.readText();
if (clipboardText && clipboardText.trim().length > 0) {
// Check for duplicates
const trimmedContent = clipboardText.trim();
const contentHash = trimmedContent.substring(0, 200) + trimmedContent.substring(Math.max(0, trimmedContent.length - 50)) + trimmedContent.length;
if (processedContent.has(contentHash)) {
console.log(`Skipping duplicate content (${clipboardText.length} chars)`);
continue;
}
processedContent.add(contentHash);
if (role === "User") {
conversation.push({
role: "User",
content: trimmedContent,
});
} else {
const processedMarkdown = processCopiedMarkdown(clipboardText, citationStyle);
conversation.push({
role: "Assistant",
content: processedMarkdown,
});
}
}
} catch (e) {
console.error(`Failed to copy from button:`, e);
}
}
}
}
// MAIN EXTRACTION ORCHESTRATOR
async function extractConversation(citationStyle) {
// Reset global citation tracking
globalCitations.reset();
// Method 1: Page-down with button clicking (most reliable)
// Uses Perplexity's native copy buttons to extract exact content
console.log("Trying Method 1: Page-down with button clicking...");
const viaButtons = await extractByPageDownClickButtons(citationStyle);
console.log(`Method 1 found ${viaButtons.length} items`);
if (viaButtons.length >= 2) {
// At least 1 complete turn (User + Assistant)
console.log("✅ Using Method 1: Button clicking extraction");
return viaButtons;
}
// Method 2: Single-pass DOM scan (no button clicking)
// Directly reads DOM content while scrolling
console.log("Trying Method 2: Single-pass DOM scan...");
const domSingle = await extractByDomScanSinglePass(citationStyle);
console.log(`Method 2 found ${domSingle.length} items`);
if (domSingle.length >= 2) {
// At least 1 complete turn (User + Assistant)
console.log("✅ Using Method 2: DOM scan extraction");
return domSingle;
}
// Method 3: Anchored copy button approach (legacy)
// Falls back to older button-based extraction
console.log("Trying Method 3: Anchored copy button approach...");
const copyButtonApproach = await extractUsingCopyButtons(citationStyle);
console.log(`Method 3 found ${copyButtonApproach.length} items`);
if (copyButtonApproach.length >= 2) {
// At least 1 complete turn (User + Assistant)
console.log("✅ Using Method 3: Anchored button extraction");
return copyButtonApproach;
}
// Method 4: Direct DOM parsing (final fallback)
// Parses visible DOM elements without any scrolling
console.log("Trying Method 4: Direct DOM parsing fallback...");
const conversation = [];
const seenUserQueries = new Set();
// Find the main thread container
const threadContainer = document.querySelector('.max-w-threadContentWidth, [class*="threadContentWidth"]');
if (threadContainer) {
// Get ALL divs in the thread container, process them in document order
const allDivs = threadContainer.querySelectorAll("div");
allDivs.forEach((div) => {
// Check if this div contains a user query (multiple possible structures)
const userQuerySelectors = [
".whitespace-pre-line.text-pretty.break-words", // Original working selector
"span[data-lexical-text='true']", // New structure selector
".group\\/query span[data-lexical-text='true']", // Grouped queries
"h1.group\\/query span[data-lexical-text='true']", // H1 queries
];
let foundUserQuery = false;
for (const selector of userQuerySelectors) {
const userQuery = div.querySelector(selector);
if (userQuery) {
const content = userQuery.textContent.trim();
if (content && content.length > 10 && !seenUserQueries.has(content)) {
seenUserQueries.add(content);
conversation.push({
role: "User",
content: content,
});
foundUserQuery = true;
break; // Stop looking once we find a query in this div
}
}
}
// If no user query found in this div, check for assistant response
if (!foundUserQuery) {
const assistantResponse = div.querySelector(".prose.text-pretty.dark\\:prose-invert");
if (assistantResponse) {
const answerContent = assistantResponse.cloneNode(true);
conversation.push({
role: "Assistant",
content: htmlToMarkdown(answerContent.innerHTML, citationStyle),
});
}
}
});
}
// Fallback: if we didn't find conversations, try the simple original method
if (conversation.length === 0) {
// Just get user queries and assistant responses in document order
document.querySelectorAll(".whitespace-pre-line.text-pretty.break-words, span[data-lexical-text='true'], .prose.text-pretty.dark\\:prose-invert").forEach((element) => {
if (element.matches(".prose.text-pretty.dark\\:prose-invert")) {
// Assistant response
const answerContent = element.cloneNode(true);
conversation.push({
role: "Assistant",
content: htmlToMarkdown(answerContent.innerHTML, citationStyle),
});
} else {
// User query
const content = element.textContent.trim();
if (content && content.length > 10 && !seenUserQueries.has(content)) {
seenUserQueries.add(content);
conversation.push({
role: "User",
content: content,
});
}
}
});
}
console.log(`Method 4 found ${conversation.length} items`);
if (conversation.length > 0) {
console.log("✅ Using Method 4: Direct DOM parsing");
} else {
console.log("❌ No content found with any method");
}
return conversation;
}
// ============================================================================
// MARKDOWN PROCESSING FUNCTIONS
// ============================================================================
// Process copied markdown and convert citations to desired style with global consolidation
function processCopiedMarkdown(markdown, citationStyle) {
// The copied format already has [N] citations and numbered URL references at bottom
// Extract the numbered references section (at bottom of each response)
const referenceMatches = markdown.match(/\[(\d+)\]\(([^)]+)\)/g) || [];
const localReferences = new Map(); // local number -> URL
// Also extract plain numbered references like "1 https://example.com"
const plainRefs = markdown.match(/^\s*(\d+)\s+(https?:\/\/[^\s\n]+)/gm) || [];
plainRefs.forEach((ref) => {
const match = ref.match(/(\d+)\s+(https?:\/\/[^\s\n]+)/);
if (match) {
localReferences.set(match[1], match[2]);
}
});
// Extract from [N](url) format citations
referenceMatches.forEach((ref) => {
const match = ref.match(/\[(\d+)\]\(([^)]+)\)/);
if (match) {
localReferences.set(match[1], match[2]);
}
});
// Remove the plain numbered references section and [N](url) citation blocks from the main content
let content = markdown
.replace(/^\s*\d+\s+https?:\/\/[^\s\n]+$/gm, "") // Remove "1 https://example.com" lines
.replace(/^\s*\[(\d+)\]\([^)]+\)$/gm, "") // Remove "[1](https://example.com)" lines
.replace(/\n{3,}/g, "\n\n"); // Clean up extra newlines left behind
// Create mapping from local citation numbers to global numbers
const localToGlobalMap = new Map();
// Build the mapping by processing all found references
localReferences.forEach((url, localNum) => {
const globalNum = globalCitations.addCitation(url);
localToGlobalMap.set(localNum, globalNum);
});
// Replace citations in the content based on the desired style
// Sort by localNum in descending order to avoid replacement interference (e.g., [10] before [1])
const sortedEntries = Array.from(localToGlobalMap.entries()).sort((a, b) => {
const numA = parseInt(a[0]);
const numB = parseInt(b[0]);
return numB - numA; // Sort in descending order
});
sortedEntries.forEach(([localNum, globalNum]) => {
if (citationStyle === CITATION_STYLES.ENDNOTES) {
// Replace [N](url) with [globalN] and [N] with [globalN]
const linkRegex = new RegExp(`\\[${localNum}\\]\\([^)]+\\)`, "g");
const plainRegex = new RegExp(`\\[${localNum}\\](?!\\()`, "g");
content = content.replace(linkRegex, `[${globalNum}]`);
content = content.replace(plainRegex, `[${globalNum}]`);
} else if (citationStyle === CITATION_STYLES.INLINE) {
// Replace [N] with [globalN](url) and keep [N](url) as [globalN](url)
const url = localReferences.get(localNum);
const linkRegex = new RegExp(`\\[${localNum}\\]\\([^)]+\\)`, "g");
const plainRegex = new RegExp(`\\[${localNum}\\](?!\\()`, "g");
content = content.replace(linkRegex, `[${globalNum}](${url})`);
content = content.replace(plainRegex, `[${globalNum}](${url})`);
} else if (citationStyle === CITATION_STYLES.PARENTHESIZED) {
// Replace with ([globalN](url))
const url = localReferences.get(localNum);
const linkRegex = new RegExp(`\\[${localNum}\\]\\([^)]+\\)`, "g");
const plainRegex = new RegExp(`\\[${localNum}\\](?!\\()`, "g");
content = content.replace(linkRegex, `([${globalNum}](${url}))`);
content = content.replace(plainRegex, `([${globalNum}](${url}))`);
} else if (citationStyle === CITATION_STYLES.NAMED) {
// Replace with ([domain](url))
const url = localReferences.get(localNum);
const domain = extractDomainName(url) || "source";
const linkRegex = new RegExp(`\\[${localNum}\\]\\([^)]+\\)`, "g");
const plainRegex = new RegExp(`\\[${localNum}\\](?!\\()`, "g");
content = content.replace(linkRegex, `([${domain}](${url}))`);
content = content.replace(plainRegex, `([${domain}](${url}))`);
}
});
// Citations processed and remapped to global numbers
// Clean up any extra parentheses that might have been created
content = content.replace(/\)\)\)/g, "))"); // Fix triple parentheses
content = content.replace(/\(\(\(/g, "(("); // Fix triple opening parentheses
// Ensure content ends with single newline and clean up extra whitespace
content = content.trim();
return content;
}
// Convert HTML content to markdown
function htmlToMarkdown(html, citationStyle = CITATION_STYLES.PARENTHESIZED) {
const tempDiv = document.createElement("div");
tempDiv.innerHTML = html;
tempDiv.querySelectorAll("code").forEach((codeElem) => {
if (codeElem.style.whiteSpace && codeElem.style.whiteSpace.includes("pre-wrap")) {
if (codeElem.parentElement.tagName.toLowerCase() !== "pre") {
const pre = document.createElement("pre");
let language = "";
const prevDiv = codeElem.closest("div.pr-lg")?.previousElementSibling;
if (prevDiv) {
const langDiv = prevDiv.querySelector(".text-text-200");
if (langDiv) {
language = langDiv.textContent.trim().toLowerCase();
langDiv.remove();
}
}
pre.dataset.language = language;
pre.innerHTML = "<code>" + codeElem.innerHTML + "</code>";
codeElem.parentNode.replaceChild(pre, codeElem);
}
}
});
// Process citations - updated for new structure with proper URL-based tracking
const citations = [
...tempDiv.querySelectorAll("a.citation"), // Old structure
...tempDiv.querySelectorAll(".citation.inline"), // New structure
];
// Track unique sources by normalized URL
const urlToNumber = new Map(); // normalized URL -> citation number
const citationRefs = new Map(); // citation number -> {href, sourceName, normalizedUrl, multipleUrls}
let nextCitationNumber = 1;
// Process citations synchronously first, then handle multi-citations
citations.forEach((citation) => {
let href = null;
let sourceName = null;
let isMultiCitation = false;
// Handle old structure (a.citation)
if (citation.tagName === "A" && citation.classList.contains("citation")) {
href = citation.getAttribute("href");
}
// Handle new structure (.citation.inline)
else if (citation.classList.contains("citation") && citation.classList.contains("inline")) {
// Get source name from aria-label or nested text
const ariaLabel = citation.getAttribute("aria-label");
if (ariaLabel) {
sourceName = extractSourceName(ariaLabel);
}
// If no source name from aria-label, try to find it in nested elements
if (!sourceName) {
const numberSpan = citation.querySelector('.text-3xs, [class*="text-3xs"]');
if (numberSpan) {
const spanText = numberSpan.textContent;
sourceName = extractSourceName(spanText);
// Check if this is a multi-citation (has +N format)
isMultiCitation = /\+\d+$/.test(spanText.trim());
}
}
// Get href from nested anchor
const nestedAnchor = citation.querySelector("a[href]");
href = nestedAnchor ? nestedAnchor.getAttribute("href") : null;
// For multi-citations, we'll process them later to avoid blocking
if (isMultiCitation) {
citation.setAttribute("data-is-multi-citation", "true");
}
}
if (href) {
const normalizedUrl = normalizeUrl(href);
// Check if we've seen this URL before
if (!urlToNumber.has(normalizedUrl)) {
// New URL - assign next available number
urlToNumber.set(normalizedUrl, nextCitationNumber);
citationRefs.set(nextCitationNumber, {
href,
sourceName,
normalizedUrl,
isMultiCitation,
});
nextCitationNumber++;
}
// If we've seen this URL before, we'll reuse the existing number
}
});
// Clean up citations based on style using URL-based numbering
tempDiv.querySelectorAll(".citation").forEach((el) => {
let href = null;
let sourceName = null;
let isMultiCitation = false;
// Handle old structure (a.citation)
if (el.tagName === "A" && el.classList.contains("citation")) {
href = el.getAttribute("href");
}
// Handle new structure (.citation.inline)
else if (el.classList.contains("citation") && el.classList.contains("inline")) {
// Get source name from aria-label or nested text
const ariaLabel = el.getAttribute("aria-label");
if (ariaLabel) {
sourceName = extractSourceName(ariaLabel);
}
if (!sourceName) {
const numberSpan = el.querySelector('.text-3xs, [class*="text-3xs"]');
if (numberSpan) {
const spanText = numberSpan.textContent;
sourceName = extractSourceName(spanText);
isMultiCitation = /\+\d+$/.test(spanText.trim());
}
}
// Get href from nested anchor
const nestedAnchor = el.querySelector("a[href]");
href = nestedAnchor ? nestedAnchor.getAttribute("href") : null;
}
if (href) {
const normalizedUrl = normalizeUrl(href);
const number = urlToNumber.get(normalizedUrl);
if (number) {
// For multi-citations, we'll show a note about multiple sources
let citationText = "";
let citationUrl = href;
if (isMultiCitation) {
// Extract the count from the +N format
const numberSpan = el.querySelector('.text-3xs, [class*="text-3xs"]');
const countMatch = numberSpan ? numberSpan.textContent.match(/\+(\d+)$/) : null;
const count = countMatch ? parseInt(countMatch[1]) : 2;
if (citationStyle === CITATION_STYLES.NAMED && sourceName) {
citationText = ` [${sourceName} +${count} more](${citationUrl}) `;
} else {
citationText = ` [${number} +${count} more](${citationUrl}) `;
}
} else {
// Single citation - use normal format
if (citationStyle === CITATION_STYLES.INLINE) {
citationText = ` [${number}](${citationUrl}) `;
} else if (citationStyle === CITATION_STYLES.PARENTHESIZED) {
citationText = ` ([${number}](${citationUrl})) `;
} else if (citationStyle === CITATION_STYLES.NAMED && sourceName) {
citationText = ` [${sourceName}](${citationUrl}) `;
} else {
citationText = ` [${number}] `;
}
}
el.replaceWith(citationText);
} else {
// Fallback if we can't find the number
}
} else {
// Fallback if we can't parse properly
}
});
// Convert strong sections to headers and clean up content
let text = tempDiv.innerHTML;
// Basic HTML conversion
text = text
.replace(/<h1[^>]*>([\s\S]*?)<\/h1>/g, "# $1")
.replace(/<h2[^>]*>([\s\S]*?)<\/h2>/g, "## $1")
.replace(/<h3[^>]*>([\s\S]*?)<\/h3>/g, "### $1")
.replace(/<h4[^>]*>([\s\S]*?)<\/h4>/g, "#### $1")
.replace(/<h5[^>]*>([\s\S]*?)<\/h5>/g, "##### $1")
.replace(/<h6[^>]*>([\s\S]*?)<\/h6>/g, "###### $1")
.replace(/<p[^>]*>([\s\S]*?)<\/p>/g, (_, content) => {
const prefs = getPreferences();
return prefs.addExtraNewlines ? `${content}\n\n` : `${content}\n`;
})
.replace(/<br\s*\/?>(?!\n)/g, () => {
const prefs = getPreferences();
return prefs.addExtraNewlines ? "\n\n" : "\n";
})
.replace(/<strong>([\s\S]*?)<\/strong>/g, "**$1**")
.replace(/<em>([\s\S]*?)<\/em>/g, "*$1*")
.replace(/<ul[^>]*>([\s\S]*?)<\/ul>/g, (_, content) => {
const prefs = getPreferences();
return prefs.addExtraNewlines ? `${content}\n\n` : `${content}\n`;
})
.replace(/<li[^>]*>([\s\S]*?)<\/li>/g, (_, content) => {
const prefs = getPreferences();
return prefs.addExtraNewlines ? ` - ${content}\n\n` : ` - ${content}\n`;
});
// Handle tables before removing remaining HTML
text = text.replace(/<table[^>]*>([\s\S]*?)<\/table>/g, (match) => {
const tableDiv = document.createElement("div");
tableDiv.innerHTML = match;
const rows = [];
// Process header rows
const headerRows = tableDiv.querySelectorAll("thead tr");
if (headerRows.length > 0) {
headerRows.forEach((row) => {
const cells = [...row.querySelectorAll("th, td")].map((cell) => cell.textContent.trim() || " ");
if (cells.length > 0) {
rows.push(`| ${cells.join(" | ")} |`);
// Add separator row after headers
rows.push(`| ${cells.map(() => "---").join(" | ")} |`);
}
});
}
// Process body rows
const bodyRows = tableDiv.querySelectorAll("tbody tr");
bodyRows.forEach((row) => {
const cells = [...row.querySelectorAll("td")].map((cell) => cell.textContent.trim() || " ");
if (cells.length > 0) {
rows.push(`| ${cells.join(" | ")} |`);
}
});
// Return markdown table with proper spacing
return rows.length > 0 ? `\n\n${rows.join("\n")}\n\n` : "";
});
// Continue with remaining HTML conversion
text = text
.replace(/<pre[^>]*data-language="([^"]*)"[^>]*><code>([\s\S]*?)<\/code><\/pre>/g, "```$1\n$2\n```")
.replace(/<pre><code>([\s\S]*?)<\/code><\/pre>/g, "```\n$1\n```")
.replace(/<code>(.*?)<\/code>/g, "`$1`")
.replace(/<a\s+(?:[^>]*?\s+)?href="([^"]*)"[^>]*>(.*?)<\/a>/g, "[$2]($1)")
.replace(/<[^>]+>/g, ""); // Remove any remaining HTML tags
// Clean up whitespace
// Convert bold text at start of line to h3 headers, but not if inside a list item
text = text.replace(/^(\s*)\*\*([^*\n]+)\*\*(?!.*\n\s*-)/gm, "$1### $2");
// This fixes list items where the entire text was incorrectly converted to headers
// We need to preserve both partial bold items and fully bold items
text = text.replace(/^(\s*-\s+)###\s+([^\n]+)/gm, function (_, listPrefix, content) {
// Check if the content contains bold markers
if (content.includes("**")) {
// If it already has bold markers, just remove the ### and keep the rest intact
return `${listPrefix}${content}`;
} else {
// If it doesn't have bold markers (because it was fully bold before),
// add them back (this was incorrectly converted to a header)
return `${listPrefix}**${content}**`;
}
});
// Fix list spacing (no extra newlines between items)
text = text.replace(/\n\s*-\s+/g, "\n- ");
// Ensure headers have proper spacing
text = text.replace(/([^\n])(\n#{1,3} )/g, "$1\n\n$2");
// Fix unbalanced or misplaced bold markers in list items
text = text.replace(/^(\s*-\s+.*?)(\s\*\*\s*)$/gm, "$1"); // Remove trailing ** with space before
// Fix citation and bold issues - make sure citations aren't wrapped in bold
text = text.replace(/\*\*([^*]+)(\[[0-9]+\]\([^)]+\))\s*\*\*/g, "**$1**$2");
text = text.replace(/\*\*([^*]+)(\(\[[0-9]+\]\([^)]+\)\))\s*\*\*/g, "**$1**$2");
// Fix cases where a line ends with an extra bold marker after a citation
text = text.replace(/(\[[0-9]+\]\([^)]+\))\s*\*\*/g, "$1");
text = text.replace(/(\(\[[0-9]+\]\([^)]+\)\))\s*\*\*/g, "$1");
// Clean up whitespace
text = text
.replace(/^[\s\n]+|[\s\n]+$/g, "") // Trim start and end
.replace(/\n{3,}/g, "\n\n") // Max two consecutive newlines
.replace(/^\s+/gm, "") // Remove leading spaces on each line
.replace(/[ \t]+$/gm, "") // Remove trailing spaces
.trim();
if (citationStyle === CITATION_STYLES.INLINE || citationStyle === CITATION_STYLES.PARENTHESIZED) {
// Remove extraneous space before a period: e.g. " [1](url) ." -> " [1](url)."
text = text.replace(/\s+\./g, ".");
}
// Add citations at the bottom for endnotes style
if (citationStyle === CITATION_STYLES.ENDNOTES && citationRefs.size > 0) {
text += "\n\n### Sources\n";
for (const [number, { href }] of citationRefs) {
text += `[${number}] ${href}\n`;
}
}
return text;
}
// Format the complete markdown document
function formatMarkdown(conversations) {
const title = document.title.replace(" | Perplexity", "").trim();
const timestamp = new Date().toISOString().split("T")[0];
const prefs = getPreferences();
let markdown = "---\n";
markdown += `title: ${title}\n`;
markdown += `date: ${timestamp}\n`;
markdown += `source: ${window.location.href}\n`;
markdown += "---\n\n"; // Add newline after properties
conversations.forEach((conv, index) => {
if (conv.role === "Assistant") {
// Ensure assistant content ends with single newline
let cleanContent = conv.content.trim();
// Check if content starts with a header and fix formatting
if (cleanContent.match(/^#+ /)) {
// Content starts with a header - ensure role is on separate line
if (prefs.formatStyle === FORMAT_STYLES.FULL) {
markdown += `**${conv.role}:**\n\n${cleanContent}\n\n`;
} else {
markdown += `${cleanContent}\n\n`;
}
} else {
// Normal content formatting
if (prefs.formatStyle === FORMAT_STYLES.FULL) {
markdown += `**${conv.role}:** ${cleanContent}\n\n`;
} else {
markdown += `${cleanContent}\n\n`;
}
}
// Add divider only between assistant responses, not after the last one
const nextAssistant = conversations.slice(index + 1).find((c) => c.role === "Assistant");
if (nextAssistant) {
markdown += "---\n\n";
}
} else if (conv.role === "User" && prefs.formatStyle === FORMAT_STYLES.FULL) {
markdown += `**${conv.role}:** ${conv.content.trim()}\n\n`;
markdown += "---\n\n";
}
});
// Add global citations at the end for endnotes style
if (prefs.citationStyle === CITATION_STYLES.ENDNOTES && globalCitations.citationRefs.size > 0) {
markdown += "\n\n### Sources\n";
for (const [number, { href }] of globalCitations.citationRefs) {
markdown += `\n[${number}] ${href}`;
}
markdown += "\n"; // Add final newline
}
return markdown.trim(); // Trim any trailing whitespace at the very end
}
// ============================================================================
// UI FUNCTIONS
// ============================================================================
// Download markdown file
function downloadMarkdown(content, filename) {
const blob = new Blob([content], { type: "text/markdown" });
const url = URL.createObjectURL(blob);
const a = document.createElement("a");
a.href = url;
a.download = filename;
document.body.appendChild(a);
a.click();
document.body.removeChild(a);
URL.revokeObjectURL(url);
}
// Create and add export button
function addExportButton() {
const existingButton = document.getElementById("perplexity-export-btn");
if (existingButton) {
existingButton.remove();
}
const button = document.createElement("button");
button.id = "perplexity-export-btn";
button.textContent = "Save as Markdown";
// Position button relative to main content area
const positionButton = () => {
// Multiple strategies to find the main content area
// Strategy 1: Find the thread content container
let mainContainer = document.querySelector(".max-w-threadContentWidth") || document.querySelector('[class*="threadContentWidth"]');
// Strategy 2: Find the input area's parent container
if (!mainContainer) {
const inputArea = document.querySelector("textarea[placeholder]") || document.querySelector('[role="textbox"]') || document.querySelector("form");
if (inputArea) {
// Walk up to find a container with reasonable width
let parent = inputArea.parentElement;
while (parent && parent !== document.body) {
const width = parent.getBoundingClientRect().width;
// Look for a container that's likely the main content (not full width, not too narrow)
if (width > 400 && width < window.innerWidth * 0.8) {
mainContainer = parent;
break;
}
parent = parent.parentElement;
}
}
}
// Strategy 3: Find main element
if (!mainContainer) {
mainContainer = document.querySelector("main") || document.querySelector('[role="main"]') || document.querySelector('[class*="main-content"]');
}
if (mainContainer) {
const rect = mainContainer.getBoundingClientRect();
// Calculate center and apply directly without transition
const centerX = rect.left + rect.width / 2;
// Force immediate positioning update by removing and re-adding styles
button.style.transition = "none";
button.style.left = `${centerX}px`;
button.style.transform = "translateX(-50%)";
// Re-enable transition after a moment
requestAnimationFrame(() => {
button.style.transition = "background-color 0.2s, left 0.2s";
});
// Debug positioning
console.log("Button positioned at:", centerX, "Container width:", rect.width, "Container left:", rect.left);
} else {
// Fallback to viewport center
button.style.transition = "none";
button.style.left = "50%";
button.style.transform = "translateX(-50%)";
requestAnimationFrame(() => {
button.style.transition = "background-color 0.2s, left 0.2s";
});
}
};
button.style.cssText = `
position: fixed;
bottom: 40px;
left: 50%;
transform: translateX(-50%);
padding: 4px 8px;
background-color: #30b8c6;
color: black;
border: none;
border-radius: 8px;
cursor: pointer;
font-size: 12px;
z-index: 99999;
font-weight: 600;
transition: background-color 0.2s, left 0.2s;
box-shadow: 0 2px 4px rgba(0,0,0,0.1);
`;
// Position initially and on resize - use immediate execution
positionButton();
// Multiple event listeners for maximum responsiveness
window.addEventListener("resize", () => {
console.log("Window resize detected");
positionButton();
});
window.addEventListener("orientationchange", () => {
console.log("Orientation change detected");
setTimeout(positionButton, 100);
});
// Watch for sidebar changes and layout updates with more aggressive observation
const observer = new MutationObserver((mutations) => {
console.log("DOM mutation detected:", mutations.length, "mutations");
positionButton();
});
// Observe body and html for any class/style changes
observer.observe(document.body, {
attributes: true,
attributeFilter: ["class", "style"],
subtree: false,
});
observer.observe(document.documentElement, {
attributes: true,
attributeFilter: ["class", "style"],
subtree: false,
});
// Use ResizeObserver on everything that could affect layout
if (typeof ResizeObserver !== "undefined") {
const resizeObserver = new ResizeObserver((entries) => {
console.log("ResizeObserver triggered for", entries.length, "elements");
positionButton();
});
// Observe multiple elements immediately
resizeObserver.observe(document.body);
resizeObserver.observe(document.documentElement);
// Also observe any main containers we can find right now
const containers = [document.querySelector(".max-w-threadContentWidth"), document.querySelector('[class*="threadContentWidth"]'), document.querySelector("main"), document.querySelector('[role="main"]')].filter(Boolean);
containers.forEach((container) => {
console.log("Observing container:", container);
resizeObserver.observe(container);
if (container.parentElement) {
resizeObserver.observe(container.parentElement);
}
});
}
// Also set up a periodic check as ultimate fallback
setInterval(() => {
const currentLeft = parseFloat(button.style.left) || 0;
const rect = (document.querySelector(".max-w-threadContentWidth") || document.querySelector('[class*="threadContentWidth"]') || document.querySelector("main"))?.getBoundingClientRect();
if (rect) {
const expectedX = rect.left + rect.width / 2;
// If button position is significantly off, reposition
if (Math.abs(currentLeft - expectedX) > 20) {
console.log("Periodic check: repositioning button from", currentLeft, "to", expectedX);
positionButton();
}
}
}, 2000); // Check every 2 seconds
button.addEventListener("mouseenter", () => {
button.style.backgroundColor = "#30b8c6";
});
button.addEventListener("mouseleave", () => {
button.style.backgroundColor = "#30b8c6";
});
button.addEventListener("click", async () => {
// Show loading state
const originalText = button.textContent;
button.textContent = "Exporting...";
button.disabled = true;
try {
// Ensure window is focused before starting
window.focus();
// Give user a moment to ensure focus
await new Promise((resolve) => setTimeout(resolve, 500));
const prefs = getPreferences();
const conversation = await extractConversation(prefs.citationStyle);
if (conversation.length === 0) {
alert("No conversation content found to export.");
return;
}
const title = document.title.replace(" | Perplexity", "").trim();
const safeTitle = title
.toLowerCase()
.replace(/[^a-z0-9]+/g, " ")
.replace(/^-+|-+$/g, "");
const filename = `${safeTitle}.md`;
const markdown = formatMarkdown(conversation);
downloadMarkdown(markdown, filename);
} catch (error) {
console.error("Export failed:", error);
alert("Export failed. Please try again.");
} finally {
// Restore button state
button.textContent = originalText;
button.disabled = false;
}
});
document.body.appendChild(button);
}
// Initialize the script
function init() {
const observer = new MutationObserver(() => {
if ((document.querySelector(".prose.text-pretty.dark\\:prose-invert") || document.querySelector("[class*='prose'][class*='prose-invert']") || document.querySelector("span[data-lexical-text='true']")) && !document.getElementById("perplexity-export-btn")) {
addExportButton();
}
});
observer.observe(document.body, {
childList: true,
subtree: true,
});
if (document.querySelector(".prose.text-pretty.dark\\:prose-invert") || document.querySelector("[class*='prose'][class*='prose-invert']") || document.querySelector("span[data-lexical-text='true']")) {
addExportButton();
}
}
if (document.readyState === "loading") {
document.addEventListener("DOMContentLoaded", init);
} else {
init();
}
})();