Adding documents

Open the collection and use Choose Files then Upload, or drag files straight onto the page.

What you can add

Documents, which get indexed and become searchable: PDF, Word (.docx), PowerPoint (.pptx), plain text, Markdown and images.

Data files, which get stored and computed over rather than searched: CSV, Excel (.xlsx, .xls) and Parquet. See the explanation of why those two are treated differently.

Up to 250 MB per file and 1000 documents per collection. The header shows where you stand: "Documents: 10 / 1000".

Watching it process

A document does not become searchable the instant it uploads. It goes through several stages, and the card tells you which one it is in:

Then a green Ready and the document is live. A long PDF with vision parsing on can take several minutes. You can leave the page: it carries on without you.

Reading a document card

Each row shows the file name, its size, the page count, and the number of chunks, which is how many searchable passages it was split into. A 9 page PDF becoming 106 chunks is normal and healthy. A long document that produced very few chunks usually means the text could not be extracted, which is the case for vision parsing.

When something goes wrong

A red error icon means processing failed. Delete the document and upload it again. If it fails a second time, the file is likely corrupt or password-protected.

"Document processing timed out" means it took too long. Usually a very large document with vision parsing on. Try it without vision parsing, or split it up.

"Embedding model not configured" or "Required models not configured" is not something you can fix. Your administrator has not finished setting up the organisation. Send them the message you are seeing.

Duplicates

Asteria notices when two documents have identical content and stores the bytes once, which is why the footprint may read "990.4 KB across 9 unique files" for ten documents. A note tells you how many overlap.

Both entries stay in the list, because the same content genuinely does arrive under two names sometimes. Nothing is broken and you are not charged twice for the storage. Tidy them up if you want to, or leave them.

The knowledge map

Underneath the counts you may see Knowledge map: 173 topics mapped. As documents are added, Asteria builds a picture of the topics in the collection and how they relate, which helps it answer questions that span several documents rather than sitting in one.

If it says Knowledge map out of date, click Rebuild. It takes a few minutes and the collection stays usable meanwhile. If it says knowledge maps are not set up for your team, that is an administrator setting.

Removing documents

Each row has a delete button. Deletion is immediate and permanent, and frees the storage.