Uploading Documents

A base can receive content through four paths:

• Text Block, to write or paste one item;
• File, to extract text from a document;
• Website, to discover and synchronize pages from a URL;
• Import CSV, to create multiple text items in bulk.

The first three are under Knowledge Bases → open the base → Add Item. Bulk import uses Import CSV at the top of the base itself. The bases.write capability is required; bases.view allows someone to inspect the base but not add or edit items.

Add a text block

The initial tab on the Add Item screen is Text Block.

1. Under Item Name, enter a name that identifies the subject, up to 100 characters.
2. Under Content, write or paste up to 100,000 characters. The editor supports Markdown.
3. Select Add Item.

The item is created as Pending and proceeds to extraction, chunking, and indexing. Its name does more than organize the list: it accompanies the chunks in the index and can help retrieval. Prefer Return policy — United States over Document 1.

Upload a file

On the File tab, SquadOS accepts one file per upload, up to 10 MB.

Content	Accepted extensions
PDF	.pdf
Word	.docx
Excel	.xlsx
HTML	.html, .htm
JSON	.json
CSV or TSV	.csv, .tsv
Markdown	.md, .markdown
Plain text	.txt

1. Fill in Item Name, or choose the file first to use its name automatically without the extension.
2. Drag the document onto the upload area or use the file picker.
3. Check that no size, format, or content-mismatch message appeared.
4. Select Send File.

The file is kept in private storage. The processor downloads the source, extracts its structure, splits the content, and creates the searchable index. Legacy Office files, presentations, ebooks, email messages, and images cannot be ingested directly: the UI tells you to convert the material to an accepted format or paste its text.

After extraction, the item can be opened through Edit. You can correct both Item Name and Content, including when the source was a file. Save Changes returns the item to Pending and starts a new indexing pass; you do not need to delete and upload the document again.

Caution

The file-signature check runs just after selection. If the UI says that the content does not match the extension, do not submit it: remove the selection and choose the correct file. The current product does not show a visual state while this short read finishes.

Index a website

Use the Website tab to turn public pages from a domain into base items.

1. Enter the Initial URL.
2. Choose a Sync frequency: Manual, Daily, Weekly, or Monthly.
3. If needed, open Advanced to set a name, limit the crawl by path prefix, exclude paths, and change the maximum page count. The initial value is 500.
4. Select Preview pages.
5. Review the number found, the pages remaining after exclusions, the credit cost, and the organization balance.
6. Select Confirm and start crawl.

Confirmation is unavailable when no effective pages were found or the balance cannot cover the cost. A frequency other than Manual creates future synchronizations, which can also consume credits. The website must be reachable by the crawler; pages that require JavaScript or a domain without a sitemap may not appear in the preview.

Import multiple items from CSV

Under Import CSV, the file must use:

• a semicolon (;) as the separator;
• title;content as the first row;
• one row per item;
• double quotes when content contains a semicolon or line break;
• two double quotes ("") to represent a quote inside an already quoted field.

Minimal example:

title;content
"Return window";"A return can be requested within 30 days."
"Support hours";"Monday through Friday, 8:00 AM to 6:00 PM."

The modal provides Download Example Spreadsheet and previews the first three valid items. The number on the import button reports how many items were recognized. Blank rows and rows without a name or content are excluded; check the count before confirming.

The import writes items one at a time. A failure in the middle of the file can therefore produce a partial result: successful rows stay in the base, and the modal lists errors for the other rows. Compare the list before retrying so you do not duplicate items that already succeeded.

Note

In the English UI, parts of the importer and the example spreadsheet may still appear in Portuguese. The file contract remains positional: the first column is the item name and the second is its content. Use the canonical title;content header and replace all example text with your own data.

Understand processing states

State	Meaning	What to do
Pending	The item is waiting for a processor.	Wait; editing also returns the item to this state.
Processing	The source is being extracted or its chunks are being indexed.	Do not upload the same content again.
Processed	Ingestion and index storage finished.	Test real questions with the agent.
Error	Extraction or indexing failed.	Check the source and use the reprocessing actions described in Organizing Base Items.

Processed does not guarantee that every question will find the answer. It confirms ingestion, not retrieval quality. For a useful validation:

1. connect the base under Agents → open the agent → Knowledge → Knowledge bases;
2. save the agent;
3. ask questions using the language customers actually use;
4. confirm that the retrieved passage supports the answer;
5. improve names, headings, and content when the information is indexed but hard to find.

See Connecting to Agents for the complete flow.

Prepare content the agent can use

• Give each item a specific name that matches customer vocabulary.
• Use headings to separate subjects; they accompany the indexed chunks.
• Keep each policy together with its conditions, exceptions, and effective date.
• Split long documents when they cover unrelated subjects.
• Remove revoked versions and duplicates to avoid contradictory answers.
• After an important import, validate the question-and-answer journey — not only the table status.