The Knowledge library page (sidebar: Build → Knowledge library) holds everything your agents can answer from: documents you upload, text you paste, and websites Moneypenny AI reads for you. Knowledge belongs to your workspace, not to one agent — add a source once, then assign it to as many agents as need it from each agent’s own Knowledge tab. See Giving an agent knowledge for that side.
At conversation time, an agent searches the sources it’s been assigned and answers from what it finds — so the quality of its answers follows directly from what you put here.
Three ways to add knowledge
Choose Add to Knowledge at the top of the page. (When your library is empty, the same three options appear as cards on the page itself.)
Upload document
Pick Upload document to add files — PDF, DOCX, PPTX, XLSX, CSV or TXT, up to 25 MB each. You can drag in up to 25 files at once; they upload a few at a time and the rest queue automatically. Moneypenny AI then reads and indexes each file in the background, and the library list shows when it’s ready to search.
If you upload a file that’s identical to one already in your library, Moneypenny AI tells you and doesn’t store it twice. If a file with the same name already exists but the content differs, you choose what happens: Replace (every agent using it gets the new version), Keep both, or Skip.
Scanned or photographed documents with unclear or handwritten text may not extract cleanly, which can affect answer quality. If you can, upload the original file rather than a scan.
Add text
Pick Add text to paste content straight in — opening hours, key policies, product facts, anything short and factual. Give it a title, paste the text, and it’s searchable like any other source. Unlike uploaded files, pasted text stays editable afterwards.
Add website
Pick Add website and enter the Website address. You can type it with or without https:// — moneypenny.co.uk works, and so does a starting page such as moneypenny.co.uk/help. Moneypenny AI starts at that page and follows links on the same site, reading each page it finds. If the address is mistyped or doesn’t lead to a public website, the dialog says why under the address box and nothing is saved until you fix it. Give it a Title if you want one — otherwise it’s named after the site’s domain. The site is read as soon as you add it, so you don’t have to wait for a refresh before its pages are ready.
Underneath, Keep this up to date sets how often the site is read again, so your agents pick up whatever has changed: Off, Daily, Weekly or Monthly. There’s no day or time to pick: the refresh repeats from when you chose it, so the first automatic refresh comes about a day, a week or a month after you save. When a workspace admin adds a website, it starts on Weekly — choose Off if, after that first read, you’d rather only ever refresh it by hand. When anyone else adds one, it starts on Off until an admin sets a refresh.
Saving the website again without changing how often leaves its refresh time where it was; choosing a different frequency, or switching it back on from Off, starts the count again from then. Scheduled tasks covers this and shows the day and time each refresh works out to. Only workspace admins can set one — everyone else sees it but can’t change it.
If a refresh can’t be set up or changed, the website is still added or saved, and the dialog stays open with a note saying why. Press Save to try again — after an add, the dialog becomes Edit website, so this doesn’t add the site a second time.
Advanced crawl settings holds the controls for how far the crawl reaches. It’s tucked away because most people never touch it, and it opens itself if any of these differs from its default:
- How deep to go — 1 fetches just the starting page’s links, 2 the next level, and so on.
- Most pages to fetch — a ceiling on how many pages are read. A new website starts at 40, and the highest you can set it is 500.
- Include patterns and Exclude patterns — limit the crawl to parts of the site (for example include
/docs/*, exclude /login).
Once added, click the website’s row to see every page that was read, and open any page to check its content. If the site is still being read, the list says so and fills in by itself when the read finishes. Use Fetch pages now on the row whenever the site has changed and you want it read straight away — it doesn’t change your schedule.
What each row tells you
Every source row shows more than a name:
- Which agents use it — each agent appears as a chip, so you can see a source’s reach at a glance. A source no agent uses yet says so.
- How a website keeps itself current — a line under the address says what it does and when it last did it: Refreshes weekly on Monday · last refreshed 2 hours ago. Without a schedule it reads Doesn’t refresh on its own, and Refresh is switched off once you’ve paused one. While a read waits its turn the line says so — Your first read is queued and will start soon for a new website, Refreshes weekly on Monday · queued and will start soon for a site that already refreshes, and Your refresh is queued and will start soon for one that doesn’t — then Fetching pages now while it’s under way. If the last read didn’t work, the line is tinted and says why and when: Refreshes weekly on Monday · we couldn’t reach the site (2 hours ago).
- Status — a healthy, ready source shows no status at all. You’ll see a chip only while something is happening (processing… for a document being read, waiting to start for a read that’s queued, fetching… for a website being read) or when something went wrong (failed, couldn’t fetch).
- What was read — uploads report back in plain terms. A spreadsheet’s row says things like how many sheets were read, how many rows in how many sections, and which column its current figures came from — so you know what the agent will be answering from before anyone asks it anything.
- Why an upload failed — when a file can’t be read, the reason appears on the row in plain words, with an Upload again link right there to retry once you’ve fixed it.
- Why pasted text failed — if text you add can’t be processed, it’s still kept: the row shows failed straight away, says what to do, and has an Edit text link. That opens your text in the editor — save it there to process it again, and once it works the row goes back to normal.
- Who uploaded it and when — uploaded files carry their provenance on the row.
Editing and removing sources
Uploaded files and crawled pages are read-only — to change one, use Download original from the row’s menu, edit the file, and re-upload it. Pasted text and website settings stay editable in place. Uploaded files also offer Preview extracted text, which shows exactly what Moneypenny AI read from the file.
If saving your edited text fails, a notification says so and the editor stays open with your text in it, so you can save again. Until a save goes through, your agents keep answering from the version before your edit.
To remove a source, use Delete on its row. If any agents are using it, the confirmation warns you how many will lose access — so you can’t quietly pull knowledge out from under a live agent. Deleting a website also removes any refresh schedule set on it.
What’s next
Adding a source here doesn’t change any agent by itself. Head to an agent’s Knowledge tab to assign sources to it, then test the agent to hear it answer from what you’ve added. 
