How to upload documents

Add documents to a project so the assistant can answer from them, and check that they processed.


Documents are what PhiBox answers from. Adding one makes it searchable for everybody who can open the project.

Two ways to add a document

Where it goesHow you add itWho sees it
Into a projectAdmin → your project → FilesEveryone who can open the project, in every future conversation
Into one conversationThe paperclip in the question boxOnly that conversation

This page covers the first. For the paperclip, see How to interact with the chat.

To add documents to a project, you need access to that project's Files section under Admin. If the project or its upload controls are not available to you, send the file to whoever administers the project.

Supported files

  • PDF: reports, contracts, articles, scans.
  • Word: .docx only. Legacy .doc files are not accepted.
  • Excel: .xlsx only. Legacy .xls files are not accepted.
  • CSV: a spreadsheet of records, where each row becomes something the assistant can look up. Download the template first and fill it in; it is the safest way to get the format right.

Upload into a project

  1. Click Admin in the top bar.
  2. Select your project at the top of the screen.
  3. Open the Files section from the list on the left.
  4. Click Upload document file and choose one or more files, up to 100 at a time and 32 MB each. For a spreadsheet of records, use the CSV upload instead.
  5. Fill in the details for each file. Every field except Metadata arrives already filled in, so nothing stops you uploading straight away, but two of them repay a moment's attention:
    • Title: filled in with the file name without its ending (e.g. without the ".pdf"). This is the name people see when the assistant cites the document, so it is worth typing something a colleague would recognise.
    • Description: filled in with Upload for and the file name, which tells the assistant nothing. One line about what is inside is very useful as it is what helps the assistant judge when this document is the relevant one.
    • Document ID: filled in for you. Leave it alone unless you know you want a particular name. Each ID must be unique within the project; an upload is refused if another document already uses it.
    • Publication Date: taken from the file's own last-changed date, which is not always the date the document was published. It cannot be left empty, so correct it when the two differ. It is what lets people filter by date later.
    • Metadata: the only field that is optional. Extra labels of your own, such as department, region or author, which let people narrow searches later.
  6. If you are uploading a batch, set Document ID format, Bulk Publication Date and the batch metadata once at the top rather than repeating yourself for every file. Individual files can still override them.
  7. Click upload. You will see Document added to queue successfully.

The upload settings include small help icons that explain unfamiliar fields in plain language, so you can fill in the rest as you go.

What happens after upload

Your file joins a queue and is processed in the background. You do not have to wait on the screen.

  1. Pending: waiting its turn.
  2. Processing: being read and split into passages so it can be searched.
  3. Completed: searchable. Ask a question about it.
  4. Failed: something went wrong. Open the queue and click Retry.

Click Queue in the Files section to watch progress; the button shows how many uploads are still waiting. The queue page lets you filter by status and by the date an upload was started, and retry anything that failed.

What to expect Small documents are searchable within a minute or so. Large ones, and scans that need to be read with OCR, can take several minutes. Until a file reaches Completed, questions about it will not find anything.

What PhiBox reads from your files

From a PDF it extracts the text, converts tables into a structured form it can search, picks up embedded images and charts, reads scans and handwriting with OCR, and straightens pages that were saved sideways or upside down. From a Word file it takes text, tables and embedded images. From a spreadsheet it takes the cell values from every sheet.

A few limits are worth knowing before you upload:

  • Faint ink, low-contrast scans and decorative fonts reduce accuracy. Clear, high-contrast text works best.
  • Pages photographed at an angle may not process correctly. Rotated pages are fine; skewed ones are not.
  • In spreadsheets, only the values you can see are read, not the formulas behind them, and images and charts inside the file are not extracted.
  • In Word files, a chart that is linked to an external Excel file rather than saved into the document will not be picked up.

The upload screen shows this same guidance for each file type, with an example, before you commit.

Best practices

  • Give every document a title a person would recognise. It is the name that appears when an answer cites it.
  • Write the one-line description. It helps then assistant decide when the document applies.
  • Upload the cleanest copy you have. A born-digital PDF beats a scan of a printout of the same thing.
  • Use metadata labels if you will want to filter later, such as department, year, region. They are what make Advanced Settings useful once a project grows.
  • Keep every Document ID unique. An upload is refused when its Document ID already belongs to another document. Delete the earlier document first if you intentionally want to replace it, then upload the new version.
  • Check the queue after a big batch and retry anything that failed, rather than assuming it all landed.

© 2026 Powered by PhiBox