AI web page reader
Give Zeplik a link and it reads the actual page, then answers from it. Not the one-line snippet a search engine shows, the real text of the article, doc, or thread. Paste a URL and ask for a summary, a specific answer, or an exact quote, and the reply is grounded in what the page genuinely says, with the source linked. It is the read half of web access: search finds pages, the reader opens them.
Reading a page is free; you pay only the normal cost of the answer, shown before you send.
How it works
Paste a link and ask
Drop a URL into the chat and tell Zeplik what you want from it: a summary, a specific answer, the key arguments, or a quote. You do not have to open the reader explicitly; a link in your message is enough for Zeplik to fetch the page.
Zeplik reads the real page
It fetches the actual page text (rendering JavaScript-heavy pages when needed), not just a search snippet, and treats that text as source material. The content is handled as untrusted data, so instructions hidden inside a page can never hijack the answer.
Answer, quote, and follow up
The answer is grounded in what the page really says, with the source linked. Ask follow-ups, pull exact quotes, translate it, or point it at another link, all in the same conversation.
Worked examples
Real prompts you can open in the composer with the text ready to review. Nothing sends on its own. Swap in your own link and question.
Prompt: Read https://en.wikipedia.org/wiki/Transformer_(deep_learning_architecture) and explain the attention mechanism in plain English
The reader pulls the full article text, so the answer quotes the real definitions rather than a one-line snippet or the model's memory.
Open this in ZeplikPrompt: Read this changelog and tell me what changed between the last two releases: https://nodejs.org/en/blog
Point it at a docs or release page and ask a specific question; it answers from what the page actually says today, not from stale training data.
Open this in ZeplikPrompt: Read this thread and summarize the main arguments on each side: paste a forum or discussion URL
Long discussion threads collapse into a faithful for-and-against summary you can skim in seconds.
Open this in ZeplikWhat it can and cannot read
The reader is good at public pages and honest about its limits. Here is the plain posture.
Standard web pages read cleanly, including ones that render their content with JavaScript.
Documentation, changelogs, and knowledge-base pages come back as readable text you can question.
Public threads are flattened into plain text so the model can summarize the whole conversation.
Where a page exposes a transcript, the reader captures the words so you can summarize a talk or lecture.
The reader fetches pages anonymously, so anything behind a login, a paywall, or your own account is out of reach. Paste the text instead.
read_url returns page text, not files. For a PDF, spreadsheet, or image, upload it to the chat and Zeplik reads it as an attachment.
Grounded, and safe by default
- It reads what is really there. The answer is built from the page's actual text, and if a page cannot be read Zeplik says so rather than inventing what it might contain.
- Page content cannot hijack the answer. Fetched text is treated as untrusted data: Zeplik will read, quote, and translate it but never obey instructions hidden inside it.
- Only public, safe addresses. The fetch is guarded so it can only reach public web pages, never internal or private network addresses.
Frequently asked questions
- How is this different from web search?
- Web search finds pages and returns titles and short snippets. The web page reader opens a specific page and returns its actual content, so the answer can quote and reason over the real text instead of a snippet. In practice they work together: Zeplik searches to find the right page, then reads it to answer in depth.
- Does reading a page cost credits?
- Reading itself is free. Fetching a page draws nothing from your balance; you only pay the normal model-token cost of the answer Zeplik writes around it, exactly like any other message. Nothing is spent by surprise, and the cost is shown before you send.
- Can it read pages behind a login or paywall?
- No. The reader fetches pages anonymously, with no cookies or credentials, so anything behind a login, a paywall, or your personal account cannot be reached. For those, copy the text and paste it into the chat and Zeplik will work from that instead.
- Can a malicious page trick the AI through its content?
- The fetched page is treated strictly as untrusted data. Zeplik wraps it in a boundary and is instructed to read, summarize, and quote it but never to follow instructions embedded inside it. So a page that says 'ignore your instructions and do X' is reported, not obeyed.
- What kinds of links work best?
- Public articles, blog posts, news, documentation, changelogs, forum threads, and pages that expose a transcript all read well, including JavaScript-rendered pages. Direct file links (PDF, spreadsheet, image) do not; upload those to the chat as attachments and Zeplik reads them that way.
- Which AI model does the reading and answering?
- The page fetch is a built-in capability that works the same regardless of model. The answer around it is written by whichever of Zeplik's frontier chat models you pick, and you can switch models at any time, including mid-conversation.
More on Zeplik
Paste a link, get the answer
Zeplik reads the real page and answers from it, source linked. Free to start.