--- title: "Document Search | Website and Files for Your Agent - Natalia" description: "Make your agent answer from your site and documents: additional pages, 500 pages per reading, PDF, .txt or .md up to 5 MB, reading progress." url: https://getnatalia.com/en/documentation/console/document-search lang: en --- [Home](https://getnatalia.com/en/) / [Documentation](https://getnatalia.com/en/documentation) / Document search Agents and actions # Document search Your website, your files, and what the agent reads The "Document search" action makes your agent answer from your own content, on whatever contact point it answers on (the web chat, for example). When you create the action, choose its source: a website, which Natalia reads and reads again, or your documents, which you upload yourself with no help from the Natalia team. ## A website Give the address of your site. Natalia reads the pages its [sitemap](https://getnatalia.com/en/documentation/natalia-analytics/glossary#plan-du-site) lists, the list of pages a site publishes for search engines, and reads them again at the next readings. ### When Natalia reads your site Natalia reads the site within a minute of saving the action, then once a day. From one reading to the next, a page whose modification date in the sitemap has not changed is not downloaded again, and a text that has not moved is not prepared again. A page that could not be read keeps its last version. A page that leaves the sitemap also leaves what the agent searches. ### Additional pages to read Give the agent a page the sitemap does not list. Under the site address, the optional "Additional pages to read" field takes a list of page addresses; each one is removed with a click. A page can sit on another site than the main address, for example an appointment booking hosted elsewhere. An added page is read like a page of the sitemap: the agent can search it, it is read again at the next readings, and it leaves what the agent can search at the next reading once you take it off the list. Changing the list is enough to start a new reading, without touching the site address. - **Address refused under the list.** An address is refused if it does not start with http or https, if it contains a username or a password, or if it points to a private address (an internal network). A page that redirects to a private address is not read. - **The page's own site decides.** Each added page follows the robots.txt file of its own site: a page that file forbids is not read. - **500 pages at most.** A list of more than 500 pages is refused. ### 500 pages per reading A reading reads 500 pages at most. Added pages go first and are never left out because of this limit. Without a sitemap If the site has no sitemap, or an empty one, the added pages are read all the same. If the sitemap cannot be reached, the whole reading stops, added pages included. ### What the last reading gave Under the site address, the action page says how the last reading went. Once the site has been read, it adds "The agent searches the content read on" followed by the date of that reading. - **Before the first reading.** "Natalia reads the site within a minute of saving, then every day." - **More than 500 pages.** The page says the site has more pages than Natalia reads, and that only the first ones are read. - **Pages not read.** Some pages could not be read: Natalia tries again the next day. - **No sitemap found.** Check the address: Natalia reads the pages the sitemap lists. - **Failed reading.** The sitemap could not be read, it lists no page, or Natalia could not prepare the pages for search. Natalia tries again the next day, and the agent keeps searching the content of the last successful reading. ### "Pending" or "To configure" As long as the agent has no content from the site yet, a badge sits under the action's name, at the top of its page. It disappears once a reading has succeeded. #### "Pending" The next step is Natalia's: the first reading has not happened yet, or Natalia could not prepare the pages and tries again the next day. You have nothing to do. #### "To configure" Natalia found no sitemap at this address, could not read it, or it lists no page. Check the site address. ## Follow the reading of a site Know when the agent will know the new content of your site. While a site is read, the agent page (under the action in the diagram) and the action page show "Natalia is reading the site.", a progress bar and the time left. A reading started from the console (creating the action, changing the address or the list) can be followed from that moment. ### Before the number of pages is known "Natalia is reading the site." shows alone, with no bar and no estimate. ### Before the first page is read The bar appears, with no estimate yet. ### While pages are read The bar shows "12 pages out of 40" and the estimate "About 3 minutes left". The estimate follows the average pace since the start: it is optimistic on a site that has mostly not changed. ### When the reading ends The screen updates by itself, with no reload, and shows the new result. - **Scheduled reading.** A reading that was not started from the console shows the next time you open the screen. - **Interrupted reading.** A reading that has not moved for 10 minutes no longer shows as in progress. ## Your documents Make your agent answer from your own documents. Create a "Document search" action with the "Your documents" source, then, on the action page, drag files or whole folders into the drop zone. A banner counts what was sent and what was refused. Once a document is ready, the agent the action is attached to looks for its answers in it. - **Accepted formats.** PDF, text (.txt) and Markdown (.md). Word, ODT and spreadsheets are not accepted: export them as PDF first. - **5 MB at most per file.** Refused at upload: a file over 5 MB, an empty file, a name that is too long, a format that is not accepted. - **Folders.** In a dropped folder, hidden files are ignored. ### The document list The list shows each document's name, its size in MB, the date it was added and its state. It shows 20 documents per page, can be searched by name, and puts the documents to fix first. It goes back to the first page when you reload. #### "In progress" An accepted document is prepared in the background. #### No state The document is ready: the agent can search it. #### Failure The document could not be prepared; the list says what to do. ### Fix a failure - **Scanned PDF.** A scanned PDF, an image with no readable text, ends in failure: text is not recognised inside an image. Upload a version that contains the text instead, for example a PDF exported from your word processor. - **Format or size.** A Word, ODT or spreadsheet file is refused at upload: export it as PDF. A file over 5 MB is refused too: split it into several files. ### Delete a document A document is deleted from the list. Once deleted, it never comes up in an answer again. Deleting the action, or changing its source, erases its documents. ## What the agent sees, and what it does not ### One action, its own documents Each action has its own documents: no other action searches them. The agents the action is attached to search them, on every contact point. ### One organisation, its own documents No other organisation searches your documents. Actions set up by the Natalia team Existing actions whose content the Natalia team had filled in keep answering exactly as before. That source is no longer offered when you create an action. ## Back to the documentation See how the chatbot answers from your documents and cites its source. [Discover the chatbot](https://getnatalia.com/en/custom-ai-chatbot) [Back to documentation](https://getnatalia.com/en/documentation)