> For AI agents: the complete documentation index is available at /en/llms.txt, the full documentation bundle is available at /en/llms-full.txt.

# Knowledge bases: ground answers in your files

You can leave a long document out of the composer. After it is in a **knowledge base**, the expert finds the relevant passages and lists sources under the reply that you can open.

Open **Knowledge Bases** from the **Settings** group in the sidebar. The page is marked **BETA**.


![Knowledge bases](/images/en/ui-knowledge.png "Knowledge bases")

## Try it with a file you already have

If the page asks you to contact an administrator, ask them to turn the feature on in the next section. When it is already on, do these four steps:


### Create a knowledge base

Click **Create knowledge base** and enter a name. Add a description and icon if you want them. If **100** documents is the wrong cap, set **Document limit**. Use `0` for no limit on this base. Turn on **Shared** so other people on this instance can read it. Turn on **Default enabled** and only _your_ new chats include it automatically. Everyone else still selects a shared base in chat.

### Add files

Click **Upload documents** or drag files onto the page. Supported text includes md / txt / rst / html / json / yaml, spreadsheets csv / tsv / xls / xlsx / xlsm, plus pdf / docx / pptx. Click **New file** to write Markdown or plain text, or **New folder** to organize files. To reorganize later, open the menu on a file or folder and click **Move to…**. A folder moves with everything inside it. You cannot move a folder into itself.

### Wait until the status is Ready

Each file moves from **Pending** or **Indexing** to **Ready**. The page shows upload progress. Saving an edit in the page indexes that file again.

### Ask in Conversation

In the composer, open **Knowledge bases**, select the bases for this turn, and ask. For example: “Which attachments does this handbook require for reimbursement?” When the expert used a document, sources appear under the reply. Click a source to preview the original layout, or download the file.

## When the page says to contact an administrator

An administrator opens **Knowledge Bases**, then **Knowledge base settings**.


### Enable knowledge bases

Turn on **Enable Knowledge Bases**. This switch applies to the whole instance.

### Choose how documents are indexed

Pick **Local ONNX** to index files on this machine, or **Online embedding** to use an embedding model you already set up under **Management → Models**.

### Download a local model before you save

For **Local ONNX**, download a model, then save. When the files are already on the host, Octop can load them offline. For Chinese documents, start with a recommended model such as `BAAI/bge-small-zh-v1.5`. Changing the embedding model or OCR settings rebuilds every knowledge index.

### Turn on OCR only if you need it

For images or scanned PDFs, turn on **Image and scanned-document OCR**. Choose **Local ONNX** (RapidOCR; components and the default Chinese and English models install when you save) or **Online vision model** (images and rendered scan pages go to the provider you pick).

With OCR on, you can also upload png / jpg / webp and scanned PDFs.

## Who can edit, and how much fits

- Only the owner or an administrator can edit a knowledge base.
- After you share a base, everyone signed in on this instance can read and select it.
- Each person can create up to **20** knowledge bases. Each base defaults to **100** documents, allows up to **10000**, and treats `0` as unlimited.
- Each upload follows the server file-size cap, which defaults to **100** MB.

## Next steps


[**Upload from code**](/en/api/knowledge.md)

Create a knowledge base and upload documents over HTTP

[**Citations in a reply**](/en/api/chat.md)

How the chat API includes knowledge sources
