Documents and the internet in agent answers

What it is

The agent reads files attached to the conversation (contracts, spreadsheets, scans, pictures), and searches and reads web pages by itself. The sites it used are shown under the answer. In spreadsheets, a link in a cell shows a page card.

How to use it

Attach a file to your question with the paperclip button Attach file, or drag it into the conversation window. The agent receives the file’s text together with your question.
The agent reads a long document in parts. Under Steps you can see Looking at the document structure, Structure “name”, Reading the document and Read “name”.
The agent looks at scans and pages with stamps and signatures as pictures, up to 4 pages at a time. It requests other pages itself (Preparing document pages) and can start text recognition (Starting text recognition, Text recognition queued) in Russian, English and Chinese, both simplified and traditional.
Ask it to find or check information on the internet. During the work you will see Searching the internet: …, Opening and Read: </strong>. You can give a page address right in your question.</li> <li>Under the answer, <strong>Sources · N</strong> expands the list of links with site icons. The full address is in the tooltip that appears on hover.</li> <li>In a spreadsheet, point at a cell with a link: after 0.7 s a card appears with the site icon, the site name, a picture and a description. If the cell holds a bare address, the button <strong>Use page title instead of the link</strong> replaces it with the page title; the link stays.</li> </ol> <div class="sl-heading-wrapper level-h2"><h2 id="limits">Limits</h2><a class="sl-anchor-link" href="#limits"><span aria-hidden="true" class="sl-anchor-icon"><svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor"><path d="m12.11 15.39-3.88 3.88a2.52 2.52 0 0 1-3.5 0 2.47 2.47 0 0 1 0-3.5l3.88-3.88a1 1 0 1 0-1.42-1.42l-3.88 3.89a4.48 4.48 0 0 0 6.33 6.33l3.89-3.88a1 1 0 0 0-1.42-1.42m8.58-12.08a4.49 4.49 0 0 0-6.33 0l-3.89 3.88a1 1 0 1 0 1.42 1.42l3.88-3.88a2.52 2.52 0 0 1 3.5 0 2.47 2.47 0 0 1 0 3.5l-3.88 3.88a1 1 0 0 0 0 1.42 1 1 0 0 0 1.42 0l3.88-3.89a4.49 4.49 0 0 0 0-6.33M8.83 15.17a1 1 0 0 0 .71.29 1 1 0 0 0 .71-.29l4.92-4.92a1 1 0 1 0-1.42-1.42l-4.92 4.92a1 1 0 0 0 0 1.42"></path></svg></span><span class="sr-only" data-pagefind-ignore>Section titled “Limits”</span></a></div> <ul> <li>Formats: PDF, DOCX, XLSX, CSV, text files in UTF-8 or with BOM, and PNG, JPEG and WEBP pictures. The old .doc format and archives (other than DOCX and XLSX) cannot be read: the file is stored, but no text is taken from it.</li> <li>A file can be up to 20 MB. Up to 200 000 characters of text are stored per file. Up to 60 000 characters from all recent files of the conversation go into the request to the agent; this limit goes to the most recent files first. The agent can read the rest in pieces of up to 12 000 characters.</li> <li>A PDF is parsed up to page 300. Page pictures: up to 4 at a time, up to 20 pages per request. Text recognition: up to 20 pages and up to 3 languages at a time, only for pages that have already been turned into pictures.</li> <li>The agent sees a picture attached to a question only while it answers that question. Pages prepared at its request it usually gets at the next step of the same answer (it waits for them up to 8 seconds). Recognized text usually arrives with the next question.</li> <li>A PDF scan without a text layer is marked <strong>no text</strong>, but the agent can still see its page pictures.</li> <li>A web page is read up to 20 000 characters, and a search gives up to 8 results. Pages protected against robots, and pages built by scripts, are read only if the platform administrator has connected Jina Reader or Firecrawl in the <strong>Agent</strong> → <strong>Data services</strong> settings.</li> <li>Internal network addresses are not opened, either by the agent or by a link card. A card is built only for http and https addresses.</li> </ul> <div class="sl-heading-wrapper level-h2"><h2 id="if-something-goes-wrong">If something goes wrong</h2><a class="sl-anchor-link" href="#if-something-goes-wrong"><span aria-hidden="true" class="sl-anchor-icon"><svg width="16" height="16" viewBox="0 0 24 24" fill="currentColor"><path d="m12.11 15.39-3.88 3.88a2.52 2.52 0 0 1-3.5 0 2.47 2.47 0 0 1 0-3.5l3.88-3.88a1 1 0 1 0-1.42-1.42l-3.88 3.89a4.48 4.48 0 0 0 6.33 6.33l3.89-3.88a1 1 0 0 0-1.42-1.42m8.58-12.08a4.49 4.49 0 0 0-6.33 0l-3.89 3.88a1 1 0 1 0 1.42 1.42l3.88-3.88a2.52 2.52 0 0 1 3.5 0 2.47 2.47 0 0 1 0 3.5l-3.88 3.88a1 1 0 0 0 0 1.42 1 1 0 0 0 1.42 0l3.88-3.89a4.49 4.49 0 0 0 0-6.33M8.83 15.17a1 1 0 0 0 .71.29 1 1 0 0 0 .71-.29l4.92-4.92a1 1 0 1 0-1.42-1.42l-4.92 4.92a1 1 0 0 0 0 1.42"></path></svg></span><span class="sr-only" data-pagefind-ignore>Section titled “If something goes wrong”</span></a></div> <table> <thead> <tr> <th>What you see on the screen</th> <th>Why</th> <th>What to do</th> </tr> </thead> <tbody> <tr> <td><strong>Reading the file…</strong> under a file for a long time</td> <td>The file is waiting in the parsing queue</td> <td>Send the question: the agent waits for parsing. The line under the file updates by itself when parsing is done</td> </tr> <tr> <td><strong>no text</strong> for a PDF scan</td> <td>The PDF has no text layer, only page images</td> <td>Ask as usual: the agent will look at the pages as pictures or recognize the text</td> </tr> <tr> <td><strong>the text is not readable</strong></td> <td>The format is not supported: old .doc, archive, binary file, unknown encoding</td> <td>Save the file as PDF or DOCX and attach it again</td> </tr> <tr> <td><strong>failed to read</strong></td> <td>The file is damaged, or parsing did not finish in time</td> <td>Re-save the file and attach it again</td> </tr> <tr> <td><strong>the agent will read the picture</strong></td> <td>It is a picture: it has no text, and the model looks at it itself</td> <td>Nothing, this is intended</td> </tr> <tr> <td>During the work, <strong>Document unavailable</strong> or <strong>Document part not found</strong></td> <td>The file was not sent together with the message, or the agent got the name of the part wrong</td> <td>Send the file with the question, or ask the agent to look at the structure again</td> </tr> <tr> <td><strong>Pages unavailable</strong></td> <td>Not a PDF, no such pages, more than 20 pages at once, or the queue is busy</td> <td>Ask for fewer pages, or try again later</td> </tr> <tr> <td><strong>Text recognition unavailable</strong></td> <td>The recognition queue is busy, or the pages or languages are wrong</td> <td>Try again later</td> </tr> <tr> <td><strong>Incomplete page address</strong></td> <td>The address has no http:// or https://</td> <td>Give the full address</td> </tr> <tr> <td>The agent says the page is unavailable or returned an error</td> <td>The site does not respond, leads to the internal network or blocks robots, and no readers are connected</td> <td>Open the page yourself; for the administrator: connect Jina Reader or Firecrawl</td> </tr> <tr> <td>The link card does not appear in a spreadsheet</td> <td>The site did not respond within 5 s, the page has neither a title nor a picture, or the address is not http(s)</td> <td>Open the link</td> </tr> <tr> <td>A globe instead of the site icon</td> <td>The site has no /favicon.ico</td> <td>Nothing, it does not affect the work</td> </tr> </tbody> </table> </div><footer class="sl-flex astro-45vocuzu"><div class="meta sl-flex astro-45vocuzu"><p>Last updated: <time datetime="2026-10-10T00:00:00.000Z">Oct 10, 2026</time></p></div><div class="pagination-links print:hidden astro-3eliszdn" dir="ltr"><a href="/en/files-storage/" rel="prev" class="astro-3eliszdn"><svg aria-hidden="true" class="astro-3eliszdn astro-ewyambfz" width="16" height="16" viewBox="0 0 24 24" fill="currentColor" style="--sl-icon-size: 1.5rem;"><path d="M17 11H9.41l3.3-3.29a1.004 1.004 0 1 0-1.42-1.42l-5 5a1 1 0 0 0-.21.33 1 1 0 0 0 0 .76 1 1 0 0 0 .21.33l5 5a1.002 1.002 0 0 0 1.639-.325 1 1 0 0 0-.219-1.095L9.41 13H17a1 1 0 0 0 0-2Z"/></svg><span class="astro-3eliszdn">Previous<br class="astro-3eliszdn"><span class="link-title astro-3eliszdn">File storage on the Files page</span></span></a><a href="/en/company-check/" rel="next" class="astro-3eliszdn"><svg aria-hidden="true" class="astro-3eliszdn astro-ewyambfz" width="16" height="16" viewBox="0 0 24 24" fill="currentColor" style="--sl-icon-size: 1.5rem;"><path d="M17.92 11.62a1.001 1.001 0 0 0-.21-.33l-5-5a1.003 1.003 0 1 0-1.42 1.42l3.3 3.29H7a1 1 0 0 0 0 2h7.59l-3.3 3.29a1.002 1.002 0 0 0 .325 1.639 1 1 0 0 0 1.095-.219l5-5a1 1 0 0 0 .21-.33 1 1 0 0 0 0-.76Z"/></svg><span class="astro-3eliszdn">Next<br class="astro-3eliszdn"><span class="link-title astro-3eliszdn">Company checks (Russia and China)</span></span></a></div></footer></div></div></main></div></div></div></div></body></html>