A data source is material the agent can quote: a PDF price list, a policy in a text file, or a pasted FAQ. Kylo splits each source into chunks, indexes them, and adds the best matches to the agent’s context on each turn.

​
Before you start

  • An approved agent with a running plan. Adding, re-syncing, and deleting are disabled while the plan is in a non-paying state (payment failed, canceled, or trial ended), and the page shows the billing card instead.
  • Files in a readable format. The uploader accepts any file, but only PDF files and text-like files can be indexed.
1

Open Build, then Data sources

Select Build, then Data sources. A new agent shows No sources with “Add files or a text snippet.”
2

Add files

Select Add files. In the dialog, drop files on the area labelled “Drop files here, or browse”, or select browse. The area says “Any file, up to 10 MB each”. A larger file is refused with ” is over 10 MB.” Select Add.Kylo reads the file, uploads it, and starts indexing. A confirmation reads “Source added and indexed” for one file or “N sources added” for several.
3

Or add a text snippet

Select Add text snippet, paste the text into Snippet, and select Add. The source name is the first non-empty line of the text, cut to 80 characters, so start with a clear title line such as “Opening hours”.
4

Read the status of each source

Each row shows the name, the type (File or Text), a line such as “Created Jan 5, 2026”, and a status badge:
5

Re-sync or remove a source

Open the actions menu on a row and choose View, Re-sync, or Delete. Re-sync at the top of the page re-indexes every source. Re-syncing a source whose text has not changed makes no new model calls. Delete asks “Delete this source?” before it removes the source.
Your agent can now retrieve from the sources marked Ready.

​
Limits you can see

  • 10 MB per file.
  • 50 MB of storage per agent. Select the storage control above the list to see the used share and the count of files and snippets.
  • Search and filter above the list: Search sources, Source type (All sources, File, Text), and Sort sources (Newest, Oldest). The list shows 10, 25, or 50 sources per page.

​
How the agent uses a source

On each turn Kylo retrieves up to six matching chunks, capped at 4,000 characters in total, and places them under the instructions as a knowledge block marked with each source’s name. The seeded instructions tell the agent to use a matching piece instead of guessing, to say so when the answer is not there, and to name the source the way a person would, for example “from the pricing doc”.

Next: test in the playground

Ask the agent a question your sources should answer.