Page Capture

Capture the current webpage as context for your AI conversation. After clicking 📄, the user message shows a Page Referenced label. Page content is attached to that user message's API payload, not inserted as a mid-conversation system message in chat history.

How to Use

Click the 📄 button → the system captures the page → a preview card appears → type your question and send. You can expand the preview or cancel the reference.

If you send with only a page reference and an empty input, the chat still shows only the reference label. Momo adds a default user query at API time (for example, asking the model to answer based on the referenced page) so local stacks such as LM Studio / Qwen do not fail on an empty user message.

Multi-turn Conversations and Tokens

Page content is injected into the API payload only for the latest user message that itself has Page Referenced.
Follow-up turns without clicking 📄 again send only the visible conversation text and do not resend the full page.
To ask another question with the full page attached again, click 📄 to reference the page and send a new message.
The 📄 icon in the assistant action bar appears only when the user message immediately before that reply has Page Referenced, and it opens only the page context bound to that turn.

💡 Web search is decided from your original typed input, not from the full referenced page content.

Capture Modes

Go to Settings → Page Capture to choose a capture mode. Smart Capture (Recommended) is the default for new installs and unset profiles.

Smart Capture (Recommended)

Automatically chooses a token-saving strategy by page type: articles, docs, and blogs use Markdown Smart Extraction; product pages, pricing pages, dashboards, settings pages, tables, and card-heavy pages use visible text extraction.

Markdown Smart Extraction

Uses Mozilla Readability and Turndown to convert the page into Markdown, preserving structure for easier reading. Capture clones the page first and processes only the clone, so the live site layout is not mutated.

If the Readability result clearly mismatches the current page title / H1 keywords (common on product pages and SPAs), that result is discarded and replaced with a fallback DOM plus visible text so hidden templates or unrelated article fragments are not captured.

💡 Best for Articles and long-text summarization. It reduces noise while preserving headings and paragraph structure.

Visible Text

Collects and deduplicates actually visible headings, paragraphs, lists, tables, buttons, and price / plan / feature blocks. Best for product pages, pricing pages, and app-like pages. If the trimmed result is too short, it falls back to document.body.innerText.

Full Page

Converts the full HTML into plain text and sends it to the conversation. Use this only when cleaner modes miss important content, as it costs more tokens.

Custom Mode

Use CSS selectors to specify which sections to include or exclude, capturing only the content you need.

Character Limit

A single capture is approximately 1,000–200,000 characters (depending on settings and the page).

⚠️ Long Context Costs The longer the content, the more context and API costs it may consume. Adjust the capture scope or mode as needed.

Smart Scrolling

Supports virtual scrolling and lazy-load detection, capturing as much visible or expandable content as possible.

Capture State Tracking

The system records the URL to avoid redundant captures on the same page.