> ## Documentation Index > Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt > Use this file to discover all available pages before exploring further. # Knowledge Bases > Collections of documents your agent can retrieve from. **Knowledge Bases** is the page that lists every knowledge base in your project. A knowledge base groups related documents your chat assistant will search together. Each KB carries a name, a primary-language hint for the chunker, and chunk-size settings. Open any KB to upload documents, see scope, and tweak settings. ## Page anatomy | Region | What you see | | ----------------- | ------------------------------------------------------------------------------------------------ | | Topbar breadcrumb | **Chat with Your Documents / {project} / Knowledge Bases**. | | Page header | Title **Knowledge bases** with a brand-gradient **+ New knowledge base** button on the end edge. | | List body | A grid of KB cards. | | Empty state | A dashed-border card with a **Create a knowledge base** button when the project has none. | ## KB list cards When the project has at least one KB, each row is a card. | Element | What it shows | | ------------------- | -------------------------------------------------------- | | Card title | The KB name. | | **Documents** count | How many documents the KB contains. | | **Chunks** count | How many text chunks the documents have been split into. | | **Size** | The total stored size, in MB or KB. | | Created date | A small caption like *Created 12 Apr 2026*. | | Click | Opens the KB detail page. | The list filters out session-scoped knowledge bases (the hidden buckets created when you drop a file into a chat composer mid-conversation). Those never appear here. ## Empty state When the project has no knowledge bases yet: | Element | What it says | | ------- | --------------------------------------------------------------------- | | Title | *No knowledge bases yet.* | | Body | *Create a knowledge base to upload documents your agent will search.* | | Button | **Create a knowledge base**. Brand-gradient pill. | The button opens the same dialog as **+ New knowledge base** in the page header. ## Creating a knowledge base The new-KB dialog has three regions: a name field, a description field, and a collapsible **Advanced** block. | Field | Notes | | ----------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Name** required | Plain-text input. Placeholder *e.g. HR policies*. The dialog blocks save until the field has at least one non-whitespace character. | | **Description** | Optional one-line summary. Placeholder *What lives inside this knowledge base?* | | **Advanced** (collapsed by default) | Two numeric fields: **Chunk size** and **Chunk overlap**. The block carries the helper text *Larger chunks carry more context per retrieval but fewer per document. Defaults work well for most cases.* | | **Create knowledge base** button | Brand-gradient submit. | | **Cancel** button | Closes the dialog. | When the request resolves, the new KB appears at the top of the list. ## KB detail page Click any card to open the KB. The detail page has three tabs: **Documents**, **Scope summary**, and **Settings**. ### Documents tab (default) The Documents tab is intentionally identical in shape to the AI Agents knowledge base step: same stats strip, same drag-drop zone, same file list, same toolbar. #### Stats strip | Card | What it shows | | ------------- | ---------------------------------------------------------------------------------- | | **Documents** | The count. | | **Chunks** | The total chunks across all documents. | | **Size** | Total stored size. | | **Language** | The primary language hint, with a small caption showing the engine's detected mix. | #### Upload zone A dashed-border drop area below the stats strip. | State | What you see | | --------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Idle | An upload icon plus the label *Drop files here, or browse to upload. PDF, DOCX, XLSX, CSV, JSON, TXT, MD, or images (JPG, PNG, WEBP, TIFF). Up to 10 MB per file.* | | Drag over | The zone tints to the brand color and the label changes to *Drop the files here...* | | Pending | A small list of files queued, with a **Cancel** button per file before they upload. | #### Toolbar above the file list | Element | What it does | | ------------- | -------------------------------------------------------------------- | | Search box | Filters by file name. | | Status filter | A dropdown: **All statuses**, **Ready**, **Processing**, **Failed**. | | Bulk actions | Once at least one row is selected: **Delete**, **Clear selection**. | #### File list | Column | Notes | | ------------ | ----------------------------------------------------- | | **Name** | The file name. Long names truncate. | | **Status** | A chip, see status pipeline below. | | **Size** | In MB or KB. | | **Uploaded** | Relative timestamp. | | **Actions** | Per-row menu with **Open**, **Re-train**, **Delete**. | A **Load more** button at the bottom paginates through long lists, with a small *Showing X of Y* count. #### Status pipeline | Chip | Meaning | Color | | ------------- | ------------------------------------------------ | -------- | | **Pending** | The file is queued. | Muted. | | **Parsing** | The engine is reading the file. | Amber. | | **Chunking** | The engine is splitting the content into chunks. | Amber. | | **Embedding** | The engine is computing vectors for the chunks. | Primary. | | **Ready** | The document is searchable in chats. | Emerald. | | **Failed** | The pipeline errored. | Red. | ### Scope summary tab A read-only summary of what the KB contains across every document. | Section | What it shows | | ------------- | -------------------------------------------------------------------- | | **Sheets** | Distinct sheet names found across spreadsheet documents. | | **Headings** | Distinct section headings found across structured documents. | | **Languages** | The languages detected, with a small percentage per language. | | Empty state | *The scope summary lights up once your documents finish processing.* | The same scope data feeds the Scope filter in the chat composer (see [Chat Canvas](/en/Systems/Chat-with-Documents/Chat-Canvas)). ### Settings tab | Field | Notes | | ------------------------- | ------------------------------------------------------------------------------------------------------------------------------- | | **Name** | Rename the KB. | | **Default language hint** | Dropdown: **Auto-detect**, **Arabic**, **English**, **Arabic + English**, **Other**. The change applies only to future uploads. | | **Chunk size** | Affects future uploads only. Existing chunks are not re-split. | | **Chunk overlap** | Affects future uploads only. | | **Save changes** button | Persists the new values. A toast confirms *Settings saved.* | | **Delete knowledge base** | A separate destructive section at the bottom. | ## Deleting a knowledge base Click **Delete knowledge base** in the Settings tab. | Element | What it says | | ------------------ | ----------------------------------------------------------------------------------------------------------------- | | Title | *Permanently delete this knowledge base?* | | Body | *This will delete every document, chunk, and vector for this knowledge base. This cannot be undone.* | | Confirmation input | Type **DELETE** to enable the confirm button. | | **Cancel** button | Closes the dialog. | | **Delete** button | Styled with the destructive (red) background and white text. Disabled until the input matches **DELETE** exactly. | After confirmation, the KB is removed and the page returns to the KB list. ## Common questions Pending normally moves to **Parsing** within seconds. If it stays for a long time, the file may be encrypted, password-protected, or a format the engine cannot read. Click the row's **Re-train** action to retry, or **Delete** to drop it. The engine reads embedded text. Pure scans need OCR; the engine handles common cases automatically but very low-quality scans may produce empty chunks. Re-export the PDF with embedded text or use a higher-quality scan. Open the **Scope summary** tab. If your headings or sheets are not in the list, the engine may have failed to extract them. Try **Re-train** on the document. If the content is in an image-only PDF, see the previous question. Not from this page. Delete from the source KB and re-upload to the target. KB scope is intentionally local so retrieval boundaries stay clear. The hint applies to future uploads only. Existing chunks were already produced. Delete and re-upload the document for the hint to take effect on it. Session-scoped KBs (created automatically when you drop a file into a chat composer) are hidden. They appear only inside the chat that owns them and are not for general project use. The bulk action confirms before running. If rows were unintentionally selected, they show as a count chip in the toolbar. The select-all checkbox at the top of the table is the most common slip, untick it before clicking **Delete**. Not yet. Each project owns its own KBs. Use the project switcher in the sidebar to flip projects, but the KB list resets to the new project's KBs. ## Where to go next Per-document detail, scope, tags, chunk inspector. Connect a read-only database alongside your KBs. Start a conversation grounded in your KBs. See storage, retrieval features, danger zone.