Knowledge stores: adding your own sources for the AI
How to add texts, documents and web pages as knowledge stores so that the AI writes with your facts, product names and arguments.
Topic area: Brand & channels
Topic area
Brand & channels- Create a channel in ALLMEDIA.AI
- Define target audience and content structure
- Using brand voice and knowledge stores
- Disconnect and reconnect a channel connection
- Channel objects in detail
- Understanding channel connections
- What you can publish per channel
- Understand and use brand voice
- Knowledge stores: adding your own sources for the AI
- Multiple brand voices per workspace
- Duplicate a brand voice and test variants
- Channel canvas and approval workflow
- Defining content structures
- Adding your own AI credentials
- Set up a publishing connection via webhook
Under Settings → Knowledge stores you create one knowledge store per source, as text, file or website. If you link it to a channel, the AI uses it in every generation for this channel. You can also add individual knowledge stores directly via “Context” when creating content.
This page was translated from the German original. The labels in your app are authoritative.
What is this about?
The AI doesn’t know your products, prices and arguments by itself. That is what knowledge stores are for. A knowledge store contains exactly one source: a text, a file or a web page. ALLMEDIA.AI reads in the content and indexes it. During generation, the AI searches it for matching passages and uses them as its factual basis.
The brand voice controls how a text sounds. The knowledge store controls what is factually correct. For brand voice, see Understand brand voice.
Some knowledge stores are created automatically: from the company website in the initial setup and from a Brand AI analysis. See Brand AI: the fast setup.
Step by step
- Open knowledge stores. Click Settings in the sidebar and then Knowledge stores. You see all knowledge stores of the current workspace, as a list or as a canvas.
- Create a new knowledge store. Click New Knowledge Store and give it a clear name, for example “Product data sheet 2026”.
- Choose the source type. Choose Text, File or Website:
- Text: You paste the content directly, such as FAQ, glossary or key messages.
- File: You upload a file and ALLMEDIA.AI extracts the text. PDF, DOCX, TXT and MD are read reliably. Older DOC files and CSV files are better saved as DOCX or TXT beforehand, otherwise no text is taken over.
- Website: You enter the Website URL and click Fetch. Under Update Interval you define whether the page is read in again manually, hourly, daily, weekly or monthly.
- Save and wait for indexing. After saving, the content is indexed. You see the progress in the detail view under Status. The AI can use indexed content for search and generation.
- Link to a channel. Open the channel under Content → Channels and select the matching stores in the Knowledge Store („Wissensspeicher“) building block in the channel canvas. From now on, every generation for this channel uses this knowledge. Details: Channel canvas.
- Test the result. Create a post for the channel on the Create page and check whether terms and facts from your source appear.
Controlling knowledge in a targeted way
- Per channel: The knowledge stores linked in the channel canvas apply to all posts of this channel.
- Per post: On the Create page you open Context and select individual knowledge stores for exactly this generation in the Knowledge tab.
- Which sources are used: Via the slider icon in the input field you open the Generation Settings. Under AI Knowledge Sources you switch Knowledge Stores, Published Posts and Scheduled Posts on or off as sources.
Limits and notes
- A knowledge store contains one source. For several documents, create several knowledge stores. Use Folder to keep them organized.
- With Website, the specified page is read in, not automatically the whole website. Create separate knowledge stores for important subpages.
- Each synchronization of a website uses credits. The detail view shows the cost per sync. TXT, MD and DOCX files as well as PDFs with a text layer are read without credits. Only scanned PDFs that have to be read via text recognition use credits.
- Some websites block automated access. In that case, reading in fails and the detail view states the reason.
- Knowledge stores belong to one workspace and are not visible in other workspaces.
Tips
- Several small, thematically clear knowledge stores are better than one large collective document. This makes it easier for the AI to find the right passage.
- Replace outdated knowledge: in the detail view you can swap a file via Replace file or read in a website again with Sync now.
- Check automatically created knowledge stores from Brand AI or the initial setup for correct numbers and names.
Related articles
checked against app version 2.29.1245
Your progress is only saved in this browser.