Appearance
Datasets
A dataset is a collection of documents you've uploaded. RAGBot reads and indexes every file so a chat can search inside it and answer from it. Group documents into datasets by topic, and you'll keep answers focused and easy to manage.
Create a dataset
Open the Datasets area in the sidebar (its page is headed Knowledge base) and choose New dataset. Give it a name that describes what's inside — Company handbook, Product specs, 2024 contracts — and an optional description. A dataset starts empty; you add documents to it next.
Company handbook
HR policies, benefits, and onboarding docs
12 files
Created Jan 3, 2026Updated 2 days ago
Support macros
Canned responses and troubleshooting guides
8 files
Created Feb 11, 2026Updated 5 hours ago
Product specs
Release notes and technical specifications
23 files
Created Mar 20, 2026Updated 1 week ago
The Datasets area. Each card shows what's inside and how many files it holds.
One topic per dataset
Smaller, focused datasets give sharper answers than one giant pile of everything. Split by subject — a single chat can always search several datasets at once if it needs a wider view.
Add sources
Open a dataset and choose Add source. A panel slides in with two tabs:
- Upload files — drag documents into the drop zone, or browse to select them. You can add several at once.
- Link — paste a web page or YouTube link. RAGBot detects which it is: a web page is scraped for its readable text, and a YouTube link is indexed from its transcript.
Either way, each new source uploads and then begins processing automatically — there's nothing else to press.
Company handbook
12 files·4.2 MB
Type
Name
Size
Chunks
Status
Added
pdf
employee-handbook-2024.pdf
2.1 MB
142
Indexed
Jun 10
docx
returns-policy.docx
320 KB
28
Indexed
Jun 10
pdf
benefits-overview.pdf
1.4 MB
—
Parsing
Jun 12
file
org-chart.png
280 KB
—
Failed
Jun 12
Documents are ready to use the moment processing finishes.
Supported file types
RAGBot reads the document formats people use every day:
- PDF — reports, handbooks, scanned policies, contracts (
.pdf) - Word — documents and memos (
.docx,.doc) - Plain text & Markdown — notes and exports (
.txt,.md)
Text-based documents work best. A PDF that's really a photo of a page (a scan with no selectable text) may not be readable — if you can't highlight the words in the file, an agent probably can't read them either.
When to use a URL instead of a file
Use Web URL when the canonical version lives on the web and changes rarely — a help article, a product page, a published policy. If a page sits behind a login or renders entirely from scripts, upload an exported copy (PDF or text) instead — RAGBot only reads what a public visitor would see.
Adding a YouTube video
Paste a YouTube video URL (a youtube.com/watch?v=… or youtu.be/… link). RAGBot uses the video's human-written captions when they exist; otherwise it transcribes the audio automatically. Only the transcript is stored — not the video or audio. Longer videos take longer to finish indexing. Playlists and channels aren't supported — add videos one at a time.
What "processing" and "indexing" mean
When you add a source, RAGBot doesn't just store it — it reads the text, splits it into small passages called chunks, and builds a searchable index of them. That index is what lets a chat later find the few most relevant passages out of hundreds of pages in a fraction of a second. It also auto-generates a handful of starter questions for each document (see below).
You don't have to do anything during this step. Each file shows one of three statuses next to its name:
- Parsing — RAGBot is still reading and indexing the file. It isn't searchable yet. Large PDFs take longer than short text files.
- Indexed — the document is fully processed and available to any chat that searches this dataset.
- Failed — something went wrong reading the file. Open the file's options and choose Reprocess to try again, or replace it with a cleaner copy.
How to tell when it's ready
Wait for every file you care about to read Indexed before you rely on the answers — a document still Parsing won't show up in results yet. The chunk count next to a file (for example 34) tells you how many searchable passages it produced.
In a dataset with many files, use the search box and the status filter at the top of the file list to jump straight to what you need.
Explore a document
Click any indexed file to open its detail panel. You'll see its status, size, and chunk count, plus two things RAGBot generated for you:
- Explore this document — a set of ready-made questions drawn from the text. Pick one to open a chat already grounded in that file. Hit Shuffle to see more.
- Chunk preview — a look at the actual passages RAGBot extracted, so you can confirm it read the document correctly.
bitcoin.pdf
File info
- Status
- Indexed
- Added
- Jan 3, 2009
- Chunks
- 34
- Size
- 184 KB
Explore this document
Showing 3 of 3Suggested starting points, generated from the indexed text. Pick one to open a conversation grounded in this file.
Chunk preview
Showing 2 of 34Chunk #1
A purely peer-to-peer version of electronic cash would allow online payments to be sent directly from one party to another…
Chunk #2
The network timestamps transactions by hashing them into an ongoing chain of hash-based proof-of-work…
A file’s detail panel: status and size, auto-generated questions, and a preview of the indexed chunks.
Organizing and deleting
Rename a dataset or a file whenever its contents drift. Inside a dataset you can remove a single document — handy when a file is out of date — without touching the rest. Deleting a document removes it from every answer that drew on it, so chats simply stop citing it.
Deleting an entire dataset removes all of its documents and unlinks it from any conversations that used it. Those chats keep working; they just have one fewer source to draw on.
Deleting is permanent
Removing a document or a dataset can't be undone. If you're replacing a file with a newer version, add the new one first and confirm it's Indexed before deleting the old.