> ## Documentation Index > Fetch the complete documentation index at: https://app.miyon.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Knowledge Base > Maintain text, websites, files, and FAQ collections centrally and connect them to multiple assistants. The **Knowledge Base** in Miyon contains the information your assistants should use to answer questions: for example, opening hours, services, directions, or return policies. You maintain this information centrally and select the appropriate sources for each assistant. ## One library for your account **Every knowledge source belongs to your account's shared knowledge base.** It can be connected to one or more assistants. This also applies when you create a source directly in an assistant's wizard: it is then available centrally and can be selected for other assistants. Creating a source and connecting it are two different steps: | Where do you create the source? | Where is it available afterwards? | Connection to an assistant | | :- | :- | :- | | **Knowledge Base** in the sidebar | In the account's shared library | Select it using **Manage** in the relevant wizard. | | **Behavior & Knowledge → Knowledge Base → Add** in the wizard | In the same shared library | After a successful save, the new entry is connected to this assistant. Other assistants select it using **Manage**. | **Example:** Both a reception assistant and an appointment assistant use the FAQ collection “Opening hours & directions”. The appointment assistant also receives the text source “Appointment rules”. If your opening hours change, edit the shared FAQ collection once. Both connected assistants receive the new version the next time their knowledge is loaded. > [!NOTE] > > “Shared” means within your account. Creating a source does not automatically activate it for all assistants. Each assistant has its own selection. ## Four types of knowledge sources Open **Knowledge Base** in the left sidebar. Here you create and manage sources independently of a particular assistant. ![Knowledge Base with Add text, Add website, Upload file, and FAQ](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/preview-knowledge-base-sidebar-icons.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=b36fa02dee4fd0d2a3134457700f4f39) | Type | Suitable for | Content of one entry | | :- | :- | :- | | **Text** | Short information, rules, and notices you maintain yourself | Freely editable text. | | **Website / URL** | Information from a publicly accessible website | The saved content of a single page or a website scan. | | **File** | Existing documents | Text read from a PDF, TXT, or Markdown file. | | **FAQ** | Recurring questions with clear answers | A named collection containing multiple question-and-answer pairs. | The screenshots on this page show the German interface. Labels may differ depending on your selected application language. ## Connect and disconnect sources in the wizard Open an assistant and go to **Behavior & Knowledge**. The **Knowledge Base** section shows the sources connected to this assistant. ![Knowledge Base in the wizard with the Add menu open: Text, URL, File, and FAQ](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/wizard-knowledge-base-hinzufuegen-menu.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=1a2962b64d5ea3145376ccd3a82802d2) ### Connect existing sources **1. Open Manage** Click **Manage** in the **Knowledge Base** section. The overview contains your account's sources, including those you created for other assistants. **2. Select sources** Search by name or type and check the entries you want to use. You can combine multiple sources and different types. **3. Apply your selection** Click **Apply**. The selected sources then appear in the assistant's Knowledge Base section. ![Manage Knowledge Base with a search field, checkboxes for File, FAQ, URL, and Text, and the Apply button](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/kb-manage.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=adb879af494f820ae7d407ea7093803f) ### Remove a connection In the list of connected sources, click the **broken-chain icon** on the right. Alternatively, open **Manage**, uncheck the source, and click **Apply**. ![Two connected knowledge sources with a disconnect icon on the right of each row](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/connected-KBs.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=ae73ed646855f3e1b0933a16f67554bf) **Disconnecting does not delete the source.** It remains in the shared library and connected to other assistants. You can reconnect it later. ### View and edit content Click a source's name to open its content. For files and websites, check particularly carefully that the extracted information is complete and understandable. FAQ collections open the question-and-answer editor; for Text, File, and URL sources, you can review and edit the saved text. > [!WARNING] > > Editing a source changes its shared content for all connected assistants. **Deleting it in the central knowledge base** removes the source and its connections. If you only want to stop using it for one assistant, disconnect it there. ## FAQ: many questions in one collection An FAQ collection is **one knowledge source containing multiple questions and answers**. You do not need a separate knowledge base entry for each question. Useful collections include “Opening hours & directions”, “Services & prices”, or “Appointments & cancellations”. ### Create a collection and edit its name inline Choose **FAQ** centrally or **Add → FAQ** in the wizard. The editor opens with an empty question. **The title at the top is an input field:** click directly on **New FAQ collection** or the existing name and enter the name you want. Press **Enter** or click outside the title to confirm it; press **Esc** to discard a name change you have not yet confirmed. ![FAQ editor with the inline-editable collection name, Question, Answer, the enabled Add another question switch, and the Save and Done buttons](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/neue-sammlung-modal.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=1be81469dad32040a41913cbf93f20ac) **1. Enter a question and answer** Phrase the question as a customer might ask it and enter a clear answer. Both fields must be filled in. **2. Save the question** Click **Save** within the question. The question is saved with the collection. The first save creates the new collection; in the wizard, it also connects it to the current assistant. **3. Add more questions or finish** Add further question-and-answer pairs or close the editor with **Done**. Previously saved questions are retained. For a new collection, the name is saved together with the first question. For an existing collection with no question open, a confirmed name change is saved directly. If you are currently editing a question, save the name together with that question using **Save**. ### What does “Add another question” do? The switch is **enabled by default**. It only controls the next editing step after a successful click on **Save**. | Setting | What happens after Save? | When is it useful? | | :- | :- | :- | | **On** | The current question is saved and a new, empty question opens immediately. | When you want to enter several questions in a row. | | **Off** | The current question is saved and collapses into a row. You return to the question list. | When you add or correct a single question. | The switch does **not** enable or disable a question for the assistant. All saved questions in a connected collection belong to its knowledge. You can start another question at any time with **Add question**. > [!NOTE] > > **Save** saves your input. **Done** closes the editor without additionally saving an open draft. Save a new or changed question first. **Cancel** within a question discards its unsaved changes. ### Maintain existing questions Click a collapsed question, edit the question or answer, and click **Save**. Use the trash icon to remove a question-and-answer pair. A saved collection needs at least one complete pair; if you no longer need the entire collection, delete its entry in the central knowledge base. > [!TIP] > > Each question should address one specific topic. Include all necessary details in the answer, such as weekdays, times, and exceptions. Resolve conflicting answers across connected sources. ## File: upload documents Choose **Upload file** centrally or **Add → File** in the wizard. Drag files into the upload area or click **Choose file**. ![Upload file dialog with Name, a drag-and-drop area, and the note PDF, TXT, Markdown, maximum 15 MB per file and 10 files or 50 MB total](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/file-upload.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=9f26e4b14273913a18c429ee53b95fc6) ### Supported formats and sizes | Format | File extensions | Notes | | :- | :- | :- | | **PDF** | `.pdf` | Use documents with selectable, extractable text. | | **Text** | `.txt` | Plain text, preferably saved as UTF-8. | | **Markdown** | `.md`, `.markdown` | Structured text with headings and lists. | | Limit | Maximum | | :- | :- | | Per file | **15 MB** | | Files per upload | **10** | | Total size per upload | **50 MB** | Word, Excel, CSV, and image files are not supported by this upload. Transfer the relevant content into a supported format first. For a single file, you can change the suggested name. With multiple files, **each file becomes a separate knowledge source**, named after its filename. In the wizard, successfully created sources are connected to the current assistant. Click **Save** and wait until uploading and processing finish. **100% uploaded** initially means that the transfer is complete; processing may still be running. If individual files fail, check the displayed filenames and errors. Successfully created entries are retained, so do not upload the entire batch again. ### What is used from the document? The assistant receives the **extracted text**, not the original page layout. Images, graphical tables, and scanned pages without a text layer are not reliable knowledge sources. If necessary, use OCR to make a scanned PDF searchable first, then check the imported content. In addition to file size, the amount of extracted text is checked. A document below 15 MB can therefore still be too large. Follow the displayed error message rather than a fixed PDF page count. Shorten such documents to the information the assistant needs or split them by topic. ## Text: enter information directly Choose **Add text** or **Add → Text** in the wizard. Give the source a descriptive **Name**, write or paste its **Content**, and click **Save**. ![Add text in the wizard with Name and Content fields and a note about the automatic connection](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/add-text-wizard.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=8606d4ddf87ec4dd6d05cab1dbec84c6) Text is suitable for company information, participation terms, or temporary notices, for example. Use short paragraphs and clear headings. For many individual questions and answers, an **FAQ collection** is easier to maintain. You can edit the text later by opening its entry. The change applies to every assistant using that source. ## Website / URL: import content Website sources contain a **saved snapshot** of the extracted website. The assistant does not open the page live on the internet for every question. The central **Knowledge Base → Add website** dialog offers two methods. The compact **Add → URL** form in the wizard imports a single page. To scan multiple pages, use the central knowledge base and connect the result in the wizard afterwards. ### Single page Choose **Single page**, enter a name and the specific page address, for example `example.com/oeffnungszeiten`, and click **Save**. Content is read from that URL without automatically importing the entire website. ![Add website in Single page mode with Name and URL](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/add-website-kb-sidebar-modal.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=f0d5a57366dc1a52dadfe2f67fd23be9) ### Entire website: advanced scraping Choose **Entire website** when the information you need is spread across several subpages. This advanced method discovers pages on the website and prepares the relevant content it finds. **1. Enter the name and website** Enter a name and the starting address. **2. Limit the page count** Set the desired upper limit under **Maximum pages**. You can choose between **1 and 100**. A higher number does not guarantee that this many pages will be found and processed. **3. Start the scan and review the result** Click **Start scan**. Progress is displayed in the knowledge base. After successful completion, the result is available as **one website source**. Open and review its content before connecting it to assistants. ![Advanced website scraping with Entire website mode, a Maximum pages slider, and Start scan](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/add-website-kb-sidebar-modal-gesamte-website.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=a55b594fbd3e0779a6601c26dc23ac61) A website scan creates a URL source. It does not create an editable FAQ collection with individual question-and-answer pairs. ### Update content with “Re-scrape” If your website changes, click **Re-scrape** on the existing entry. The content is read again while the entry and its connections are retained. For a source created by an entire-website scan, the advanced method runs again. ![Website source with URL, date of the last scrape, and the Re-scrape button](https://mintcdn.com/miyon/u6b7mRXAITO52yh-/images/knowledge-base/neu-scrapen-miyon.jpg?fit=max\&auto=format\&n=u6b7mRXAITO52yh-\&q=85\&s=37cf07bf33021f679eee4ada48b823be) Review the new content afterwards. If you previously added manual edits to the imported text, re-scraping may overwrite them. Maintain your own permanent additions in a separate Text or FAQ source instead. > [!NOTE] > > A website must be accessible for extraction. Sign-in requirements, access protection, or blocked automated requests can prevent an import. In that case, add the information you need as text or a supported file. Even after a successful scan, check that all important content is present. ## Keep knowledge current and manageable * **Select appropriate sources:** Connect only the knowledge that a particular assistant needs. Merely having a source in the library is not enough. * **Maintain shared content centrally:** Update the existing source instead of creating copies for every assistant. Use separate sources for information that differs, such as details for different locations. * **Test changes:** Start a new test conversation after saving. Voice assistants load the saved knowledge when a new conversation is set up; editing does not retroactively update an ongoing conversation. * **Separate knowledge from instructions:** Facts belong in the knowledge base. Describe response style, what to do when information is missing, and workflows in the [assistant's instructions](/docs/en/ai-prompts). Connected sources are provided automatically as additional context; no manual prompt placeholders are needed. * **Keep content concise:** Avoid duplicate or conflicting text. File size and text length are different limits; the application tells you when content is too extensive. ## Context budget The main instruction and the connected knowledge sources are sent to the language model together on every conversation. The wizard shows a yellow or red icon in the header of the main instruction as soon as this context reaches the recommended size. What is counted, how Standard and Realtime Conversation differ and how to stay within the budget is described under [Context budget](/docs/en/context-budget). ## Frequently asked questions > **Why doesn't my assistant know an existing source?** > > Under **Behavior & Knowledge → Knowledge Base**, check whether the source is connected to this assistant. If necessary, select it under **Manage** and click **Apply**. Also check its saved content and start a new test conversation. > **Do I need to create a source again for every assistant?** > > No. Select the same source for additional assistants under **Manage**. A source originally created in the wizard is available there too. A copy only makes sense when the content needs to differ independently. > **Can I connect only part of an FAQ collection?** > > A connection applies to the entire collection. **Add another question** only controls data entry in the editor. If assistants should receive different questions, divide them into suitable collections. > **Why is a small PDF rejected or only partially read?** > > In addition to the 15 MB per-file limit, there are limits on extracted text. PDFs also need an extractable text layer. Check the error message, test whether you can select text in the PDF, and review the imported content. OCR beforehand may help with image-based documents. > **Is a website opened again for every answer?** > > No. A URL source contains the saved import. Use **Re-scrape** to bring in website changes. If the assistant needs current data for an individual request, use a suitable [tool](/docs/en/tools). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.