> ## Documentation Index > Fetch the complete documentation index at: https://docs.tied.com/llms.txt > Use this file to discover all available pages before exploring further. # Help Center > A public knowledge base for customers, and the knowledge source your AI agent answers from. The Help Center does two jobs at once, and the second is the one people underestimate. **Publicly**, it is a knowledge base your customers can search and read — articles, categories, FAQs, on your own domain if you want. **Internally**, it is the highest-quality thing you can point an AI agent at. An agent with no knowledge source can only escalate. An agent with a well-maintained Help Center can resolve a `knowledge` question outright, which is the cheapest resolution class there is and the one most of your volume will fall into. **Writing your Help Center is agent training.** Every article you publish is something your agent can answer from without touching an external system. This is the highest-leverage content work in the product — it serves customers who prefer self-service *and* deflects the questions your agent would otherwise escalate. ## How it is structured One published knowledge base — name, slug, branding, and optionally a custom domain. Most accounts need one. Run several to serve genuinely separate brands or audiences. Groups articles. Categories **nest** — a category can have sub-categories — and can be linked to related categories for cross-referencing. Belongs to a portal, and optionally to a category. Has an author and a status: `draft`, `published` or `archived`. A portal can also be attached to a **web chat widget**, which is what puts Help Center search inside the widget so customers can find an answer before starting a conversation. ## Translations work by linking, not duplicating This is the part worth understanding before you start writing in more than one language. A translated article is not a copy. Each article has a **root** and a set of **associated** articles — its translations. Categories work the same way. The relationship is explicit, which means a reader gets routed to their language rather than to a near-duplicate that search treats as competing content. Plan the root language first and translate from it. Retrofitting a translation structure onto articles written independently in three languages is considerably more work than doing it in order. ## Article states | State | Visible publicly | Agent can answer from it | | - | - | - | | `draft` | No | No | | `published` | Yes | Yes | | `archived` | No | No | **Archiving an article removes it from your agent's knowledge.** If customers still ask about something, archiving the article does not stop the questions — it just stops the agent being able to answer them. Update rather than archive unless the topic is genuinely dead. ## Where to start Not the ones you wish you got. Pull them from your existing inbox. Ten real articles beat fifty speculative ones, both for customers and for the agent. Structure emerges from the content. Do not design a taxonomy before you have articles to put in it. A published article is immediately available as agent knowledge. See [AI agents](/agents). Every escalation your agent could not answer is a gap in your knowledge base. That feedback loop is the point. ## In this section Setup, slug, branding, and running more than one. Nesting, ordering and related categories. The editor, draft state, authorship and SEO. Question-and-answer sets, and how they differ from articles. The root-and-associated model, in practice. Theming, domain, search behaviour and visibility. ## Next Pointing an agent at the Help Center you just wrote. Human agents using the same content in the inbox. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.