Atender — Knowledge, the source of truth your customers and your AI read.

The workspace

How the content is authored

Knowledge starts from a single working assumption: most of the work in a help centre isn’t writing new articles, it’s keeping the existing ones correct. Every choice in the workspace is shaped to make ongoing maintenance fast.

1The three-pane workspace — folders, articles, editor, all in one view

Left

Folders

The category and subcategory tree. Click a folder to load its articles in the middle pane.

Middle

Articles

Every article in the selected folder, with its status, last updated, and view count visible at a glance.

Right

Editor

Full rich-text editor with AI assist. Edit one article without losing sight of the rest of the library.

One view, the whole library. The point of three panes side by side is that a maintainer can see the context an article lives in while editing it — the folder it sits in, the articles around it, and the article itself. Most help-centre tools force you to navigate away to find the next thing. Knowledge keeps the next thing visible.

A folder tree is one axis. Knowledge has three more. ↓

2Organisation beyond folders — the same article can be reached from several directions

Tags cross-cutting labels

Flexible cross-category labels

An article on resetting a password can live in Account and still be tagged Security and Getting Started — making it discoverable from any of the three.

Sections rule-based collections

Auto-filling collections of articles

Define a content rule (by category, tag, or specific articles) and the section auto-fills, sorts itself by newest, oldest, or alphabetical, and gets its own hero on the public site.

Roles optional perspectives

Audience-specific views of the same library

Build admin, team-lead, and end-user views and let visitors switch between them with a chip in the header. Optional — many tenants never turn this on.

Bulk actions across many articles. Select a batch and publish, unpublish, archive, restore, retag, or change role visibility in one move — useful when a product change ripples across dozens of articles at once.

The public site

How the public help site is built

Your public help site isn’t a fixed template. It’s a drag-and-drop canvas. Pick a layout template as the starting structure, drop in the blocks you want, paint the brand on top — without writing a line of code.

1Layout templates — the starting structure of the site

Classic

Sidebar

Spotlight

A centered, conventional help-centre layout. Hero on top, categories below, footer underneath. Familiar to every visitor.

Persistent navigation rail on the left. Best for documentation-heavy sites where visitors browse a deep tree.

A marketing-style landing with a large hero. Best when most visitors are new and arrive via search.

Then arrange the blocks. Sixteen of them. ↓

2The block library, in context — each block does one thing; you arrange them on the canvas

Sixteen blocks in total — HeroSearch and CompactSearch for discovery; CategoryGrid / CategoryList / SectionGrid / SectionList / SectionBar for navigation; FeaturedArticle / RecentArticles / WhatsNew for content surfaces; Announcement / QuickLinks / ContactCTA / VideoEmbed / Spacer for chrome and engagement; and SystemStatus wired into the Incidents module. Drop them on the canvas in any order. The active layout template, theme, and language all adapt to whatever you build.

The site isn’t isolated from the rest of Atender. The SystemStatus block reads live state from the Incidents module — when there’s an active incident, your help site says so without anyone having to update it. Block-level cross-module links like this are the long game.

Then paint the brand on top ↓

3Theme and branding — configured once, applied everywhere on the site

Light & dark independent palettes

Two complete colour configurations

Primary, secondary, background, text, and accent — configured separately for light and dark mode. The visitor’s system preference picks the right one.

Identity logo, favicon, social

The marks of your brand

Upload a logo and favicon, set custom fonts, drop in social profile links — placed automatically in header, footer, and metadata.

Chrome header, footer, sidebar

Every chrome element is configurable

Header links, footer content, sidebar navigation, breadcrumbs — all editable. Configure once, applied across every page on the site.

Navigation how visitors move

Choose how customers browse

Persistent sidebar with sections or categories, header links, or no chrome at all. Pick the model that matches the depth of your library.

Reach & intelligence

One source, every market, every language

This is what makes Knowledge multiplicative. You write articles in your own language. Atender serves the right version to every visitor in their own language and for their own market — and the same library powers Agent Stacks behind the scenes.

1Write once — translated automatically — translation isn’t a task on someone’s plate

How it runs

Background, continuous, hands-off

When you publish or update an article, Atender translates it into every active language automatically. You don’t queue jobs. You don’t wait. You don’t run scripts. Each language has a state — Active, Pending, or Disabled — and a translation pipeline that tracks every job, shows per-language progress, and flags failures so a human can review them.

Translation is a property of the system, not a task on someone’s plate. Update an article in English at 14:32, and the German, French, Spanish, and Japanese versions catch up on their own. There’s no “translation day,” no batched release.

And it goes one step further than translation. ↓

2Market-specific content inside one article — same article, different bodies, no duplication

Europe

230V

United Kingdom

230V

United States

120V

One article about electrical sockets — three different visitor experiences. Each region sees the right voltage and the right plug shape. Same article underneath, no duplication.

How authors do this. Authors drop Market Content and Market Image blocks inside the article and assign each one to a country, region, or fall back to global. Atender picks the best variant by walking the hierarchy — country → region → global. So a variant set for Europe is shown to every European country that doesn’t have a more specific variant of its own.

Some words shouldn’t be translated. Some should — very specifically. ↓

3Protected terms — a name in your product is a name, not a phrase to translate

Never translate brand names, products

Keep the original in every language

Brand names, product names, and feature names that should always stay in the original. My Page stays My Page in German, Spanish, and Japanese.

Per-language overrides curated translations

One specific translation per language

The exact translation, defined per language. My Page becomes kundesenter in Norwegian, Mi Página in Spanish, マイページ in Japanese — every time, in every article.

Matching can be case-sensitive or case-insensitive. The same protected-terms list is shared with Auto-Translation on the conversation side, so the rules stay consistent across the whole product — help site, chat, email, voice.

How does the right version reach the right visitor? ↓

4Detection and override — the visitor lands on the right version automatically

Automatic browser language

Detected from the visitor’s browser

Knowledge reads the visitor’s browser language settings and serves the matching language version on first paint.

Manual selector in the header

The visitor can switch any time

A market-and-language selector sits in the header. Useful when the visitor is travelling, on a VPN, or wants to see content for a different region.

Under the hood

Your help site is your AI’s training corpus, automatically

Two things about how Knowledge works underneath are worth saying out loud, because they shape what the rest of the product can do with it.

Vectorized pgvector + HNSW

Every article is chunked, embedded, and indexed

Articles are split section-first by heading, embedded into a vector index, and stored alongside the original content. When Agent Stacks needs an answer, it does semantic retrieval — pulling the chunks that match the customer’s intent, regardless of whether the customer used the same words you did. The same index also powers visitor search inside the public help site.

Continuous no scheduled batches

Translations and embeddings catch up on their own

There’s no “translation day,” no nightly job to babysit. The translation pipeline runs in the background as content changes. The embedding pipeline does the same — the moment an article is updated, it’s re-chunked and re-embedded so semantic search and Agent Stacks stay current.

This is the dual-purpose payoff of one library. Every article you write does the same job twice — it answers customers directly through your help site, and it teaches Agent Stacks how to answer the customers who never read it themselves. Maintain one corpus, get two channels.