Articles & File Uploads
View as Markdown
Inside each [Collection](/knowledge-base/collections/), you can add knowledge directly in two ways: **authoring articles manually** in the built-in editor or **uploading document files** for automatic background extraction and vector indexing.
:::note[Required permission]
You need **`ai:manage`** or **`workspace:manage`** to create, edit, or delete knowledge sources.
:::
## Knowledge Source Kinds
When viewing or creating sources under **Articles** (`/{slug}/ai-kb/{collectionId}/articles`), each source carries a type:
| Kind | Purpose | Directly Editable |
|---|---|---|
| **Article (`ARTICLE`)** | Guides, workflows, and detailed context for AI responses | Yes |
| **Exact Match (`EXACT_MATCH`)** | Fixed rules that return a verbatim response without LLM rephrasing | Yes |
| **FAQ (`FAQ`)** | Concise question-and-answer pairs for common inquiries | Yes |
| **Policy (`POLICY`)** | Return, warranty, shipping, and payment terms | Yes |
| **Product (`PRODUCT`)** | Product/service descriptions (also creatable via Mina or API) | Yes |
| **File Document (`FILE_DOC`)** | Content extracted from uploaded files | Yes (after processing) |
| **Web Page (`WEB_PAGE`)** | Content synced from a website URL | Read-only (resync from source URL) |
| **Chat Example (`EXAMPLE`)** | Reply examples harvested from past human conversations | Read-only |
## Write Articles & Exact Match Rules
1. Inside a collection, click the add button and choose **Write Article**.
2. Select a content kind (**Article**, **Exact Match**, **FAQ**, or **Policy**).
3. Enter a **Title** and compose the content in the rich-text editor (supports formatting and inline image uploads).
4. For **Exact Match (`EXACT_MATCH`)** sources, add **question variants** (`questionVariants`) so the rule triggers reliably across different phrasings.
5. Click **Save** (or press **`Ctrl+S`** / **`Cmd+S`** on the article detail view). You can toggle active status (`isActive`) or move an article to another collection at any time.
## Upload Documents (File Ingestion)
OmnyBot supports uploading files that are asynchronously parsed, chunked, and embedded via a background worker queue (`ai-document-ingest`).
### Supported Formats & Limits
- **File extensions**: `.pdf`, `.docx`, `.txt`, `.md`, `.markdown`, `.csv`, `.json`
- **Max file size**: **100 MB** per file
- **Batch limit**: up to **10 files** per upload batch
### Upload Workflow
1. Inside a collection, click the add button and choose **Upload Document**.
2. Drag and drop or browse files from your device (unsupported extensions, files over 100 MB, and duplicates in the same batch are skipped automatically).
3. Start the upload to transfer files to storage and enqueue them for background processing.
4. Track indexing status on each item:
- **`PENDING`** → **`INDEXING`**: Extracting text and generating embeddings.
- **`READY`**: Indexed and searchable by AI bots.
- **`FAILED`** / **`ERROR`**: Processing failed (inspect the recorded error message).
:::tip[Automatic chapter splitting for large documents]
Documents exceeding **25,000 words** are automatically split into child chapters (~15,000 words each) with a generated table of contents on the parent source for better retrieval accuracy. **`.csv`** files are converted row-by-row into structured `Column: Value` chunks before embedding.
:::
:::caution[Scanned image PDFs]
Text is extracted directly from the PDF text layer. Image-only scanned PDFs (or files yielding fewer than 10 characters of text) will fail ingestion.
:::
Inside each Collection, you can add knowledge directly in two ways: authoring articles manually in the built-in editor or uploading document files for automatic background extraction and vector indexing.
Knowledge Source Kinds
Section titled “Knowledge Source Kinds”When viewing or creating sources under Articles (/{slug}/ai-kb/{collectionId}/articles), each source carries a type:
| Kind | Purpose | Directly Editable |
|---|---|---|
Article (ARTICLE) | Guides, workflows, and detailed context for AI responses | Yes |
Exact Match (EXACT_MATCH) | Fixed rules that return a verbatim response without LLM rephrasing | Yes |
FAQ (FAQ) | Concise question-and-answer pairs for common inquiries | Yes |
Policy (POLICY) | Return, warranty, shipping, and payment terms | Yes |
Product (PRODUCT) | Product/service descriptions (also creatable via Mina or API) | Yes |
File Document (FILE_DOC) | Content extracted from uploaded files | Yes (after processing) |
Web Page (WEB_PAGE) | Content synced from a website URL | Read-only (resync from source URL) |
Chat Example (EXAMPLE) | Reply examples harvested from past human conversations | Read-only |
Write Articles & Exact Match Rules
Section titled “Write Articles & Exact Match Rules”- Inside a collection, click the add button and choose Write Article.
- Select a content kind (Article, Exact Match, FAQ, or Policy).
- Enter a Title and compose the content in the rich-text editor (supports formatting and inline image uploads).
- For Exact Match (
EXACT_MATCH) sources, add question variants (questionVariants) so the rule triggers reliably across different phrasings. - Click Save (or press
Ctrl+S/Cmd+Son the article detail view). You can toggle active status (isActive) or move an article to another collection at any time.
Upload Documents (File Ingestion)
Section titled “Upload Documents (File Ingestion)”OmnyBot supports uploading files that are asynchronously parsed, chunked, and embedded via a background worker queue (ai-document-ingest).
Supported Formats & Limits
Section titled “Supported Formats & Limits”- File extensions:
.pdf,.docx,.txt,.md,.markdown,.csv,.json - Max file size: 100 MB per file
- Batch limit: up to 10 files per upload batch
Upload Workflow
Section titled “Upload Workflow”- Inside a collection, click the add button and choose Upload Document.
- Drag and drop or browse files from your device (unsupported extensions, files over 100 MB, and duplicates in the same batch are skipped automatically).
- Start the upload to transfer files to storage and enqueue them for background processing.
- Track indexing status on each item:
PENDING→INDEXING: Extracting text and generating embeddings.READY: Indexed and searchable by AI bots.FAILED/ERROR: Processing failed (inspect the recorded error message).