> ## Documentation Index
> Fetch the complete documentation index at: https://docs.serval.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Authoring Knowledge in Serval

> Write and publish knowledge directly in Serval with a block editor, publishing workflow, and built-in agent retrieval

You can author knowledge bases directly in Serval — no external wiki required. You write pages in a rich block editor, publish them when they're ready, and the help desk agent searches and cites them like any other knowledge source.

Knowledge bases you author live in the **Knowledge Bases** tab of the Knowledge section, alongside the **Knowledge Sources** tab that manages synced content.

## Creating a knowledge base

1. In the sidebar, open **Knowledge**, then select the **Knowledge Bases** tab
2. Click **New knowledge base** and give it a name (for example, "IT Handbook")
3. Add pages and folders to build out your content

Each knowledge base has a name and an icon you can customize. Your team can create as many knowledge bases as it needs — for example, one per department or per topic area.

### Pages and folders

Content is organized as a tree:

* **Pages** are documents you write in the block editor. Pages can nest under other pages, so a page is also a container for its children.
* **Folders** are pure grouping containers for pages.

Drag items in the sidebar to reorder or move them.

## Writing content

Pages use a block editor with the formatting you'd expect from a modern wiki:

* Paragraphs and headings (H1–H3)
* Bulleted, numbered, and checklists
* Quotes, code blocks, and dividers
* Tables
* Images — paste, drag-and-drop, resize, and caption

Two inline shortcuts speed up writing:

* Type **`[[`** to link to another page in the same knowledge base
* Type **`@`** to mention a teammate

Everything autosaves as you type — a "Saving…" indicator appears while a write is in flight.

## Drafts, publishing, and versions

Pages follow a **draft → publish** model:

| State                                                     | Who sees it          | Agent searches it? |
| :-------------------------------------------------------- | :------------------- | :----------------- |
| **Draft** (never published, or edited since last publish) | Editors only         | No                 |
| **Published**                                             | Everyone with access | Yes                |

* Every page starts as a draft. **Only published pages are indexed for agent retrieval** — you can iterate freely without half-finished content reaching users.
* Clicking **Publish** (or **Publish changes**) snapshots the current content as a new version. The version pill in the header (for example, `v3`) shows a green dot when the live version is published and current, and an amber dot when there are unpublished changes.
* **Version history** shows the timeline of published revisions — autosaves don't create versions, so history stays meaningful.
* **Unpublish** retracts a page: the agent stops searching it and read-only viewers lose access.

After you publish, an **indexing status** badge shows where the page is in the search pipeline: **Synced & searchable**, **Indexing**, or **Index failed** (with a retry option for publishers).

## Organizing and finding content

* **Search** — full-text search across pages from the sidebar
* **Favorites** — star pages for yourself; favorites are personal
* **Pinned** — pin pages for the whole team; everyone sees the same pinned set
* **Archive** — archiving moves a page to Trash, where it can be restored; deleting is permanent

Pages also support **duplicate**, **move** (to another parent or knowledge base), and **copy link** from the page actions menu.

## Access control

Knowledge bases authored in Serval have two independent layers of access control:

<Info>
  **Workspace access** controls who can view and edit pages inside Serval. **Help desk audience** controls who the agent surfaces a page to when answering requests.
</Info>

### Workspace access

Team roles govern who can view, create, edit, publish, archive, and delete authored knowledge content. Members with read-only access see published pages with a read-only banner — they never see drafts or version history.

### Help desk audience

Each page has a **Visibility** control ("Who can find this in the help desk") that scopes which users the agent will surface the page to:

* Leave it empty for **anyone in the org**
* Select access profiles to restrict the audience

Child pages inherit their parent's audience automatically. Restricting the help desk audience does **not** change who can view or edit the page in the workspace — the two layers are separate.

### AI-context-only knowledge bases

A knowledge base can be marked **AI context only**: its content is indexed for agent retrieval but hidden from human-facing surfaces. Use this for internal reference material the agent should know without exposing it as browsable documentation.

## How the agent uses authored knowledge

When a user submits a request, the help desk agent searches your published pages together with all other knowledge sources. Answers cite the page with a **View in Serval** link that deep-links straight to it.

## Learn more

* [Knowledge Base overview](/sections/documentation/knowledge-base/overview) — all the ways to get knowledge into Serval
* [Connect to External Sources](/sections/documentation/knowledge-base/connected-sources) — sync from your documentation apps
* [Building Custom Ingestion Workflows](/sections/documentation/knowledge-base/custom-ingestion) — pull knowledge from systems without a first-party connector
* [Skills vs. Knowledge](/sections/documentation/skills-vs-knowledge) — what belongs in knowledge vs. a Skill