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

# Overview

RAUM **AI Chatbot** is the customer-facing website support product.

It uses the same organization knowledge, model configuration, and connection infrastructure as AI Assistant where those resources are compatible, but it has a separate public-access boundary. Internal Zendesk context is not automatically exposed to website visitors.

## What Chatbot can do

A published website widget can:

* answer from Chatbot-approved procedures, guidelines, documents, websites, Gallery assets, Updates, and authored public knowledge;
* use the AI provider and models configured for your organization;
* retrieve approved live data from bounded public SOURCE connections;
* perform explicitly configured public ACTION workflows with identity, confirmation, input bounds, rate limits, idempotency, and audit controls;
* accept supported visitor attachments;
* support multiple published interface languages and organization-defined welcome messages;
* collect pre-chat information;
* offer a human-handoff form on configured triggers;
* create internal RAUM requests, send handoff email, or create Zendesk tickets;
* collect CSAT;
* classify topics, sentiment, visitor context, and deterministic external entities when those features are enabled; and
* provide conversation, handoff, action, topic, visitor, cost, and widget analytics.

## Chatbot is public by default only where you approve it

The most important product boundary is simple:

> [!WARNING]
>
> A resource that AI Assistant can use is **not automatically available to AI Chatbot**.

Chatbot publication is separately reviewed because the visitor is outside your support organization.

**Library → Shared Knowledge** is the central public review surface. Administrators approve resources for Chatbot there.

Public resource types include:

* Procedures
* AI Guidelines
* Documents & Websites
* Gallery assets
* Connections
* MCP Servers
* Updates
* RAUM-authored articles

Zendesk fields, macros, groups, organizations, and similar internal Zendesk metadata are not copied into public Chatbot context.

## Product boundaries

Some Assistant features are intentionally not Chatbot features.

| Feature | AI Assistant | AI Chatbot |
| - | - | - |
| Procedures | Yes | Yes, after public approval |
| Response AI Guidelines | Yes | Yes, after public approval |
| Translate-specific Guidelines | Yes | No |
| Documents & Websites | Yes | Yes, after public approval |
| Zendesk Help Center synchronization | Yes | Not automatically public |
| Zendesk Ticket Fields | Yes | No public context |
| Zendesk Macros | Yes | No public context |
| Brand Voice | Yes | No |
| AI Coaching | Yes | No |
| API/MCP connections | Internal use | Only through explicit public policy + approval |
| Public visitor identity | No | Yes |

## Main administration areas

> **[Setup](https://docs.raum.am/chatbot/setup.md)**
>
> Complete the minimum path from AI model to live test widget.

> **[Shared Knowledge](https://docs.raum.am/chatbot/shared-knowledge.md)**
>
> Review exactly what is allowed to become public.

> **[Channels](https://docs.raum.am/chatbot/channels.md)**
>
> Configure exact website origins, channel state, publishable keys, and optional handoff delivery.

> **[Widgets](https://docs.raum.am/chatbot/widgets.md)**
>
> Build, publish, localize, install, and preview versioned website widgets.

> **[Forms & Handoffs](https://docs.raum.am/chatbot/forms-handoffs.md)**
>
> Collect pre-chat data and route human-support requests.

> **[Conversations](https://docs.raum.am/chatbot/conversations-topics.md)**
>
> Review retained conversations, failures, exports, legal holds, and deletion.

> **[Topics](https://docs.raum.am/chatbot/topics.md)**
>
> Classify conversation intent and sentiment and control visitor-data analytics.

> **[Analytics](https://docs.raum.am/chatbot/analytics.md)**
>
> Measure containment, handoffs, CSAT, actions, language, topics, widgets, and AI cost.

## Recommended launch sequence

1. Connect and verify an AI provider.
2. Review the knowledge you intend to expose publicly.
3. Create a web channel with exact production origins.
4. Create a widget draft.
5. Configure languages, appearance, attachments, forms, knowledge collections, and CSAT.
6. Review public connections/actions and visitor-identity requirements.
7. Publish the widget.
8. Copy the one-time publishable key and installation snippet.
9. Test with Preview and private Simulation.
10. Run representative live tests on an allowed origin.
11. Monitor Conversations, Requests, Analytics, Topics, and failed deliveries after launch.

Chatbots are deceptively good at looking finished once a bubble appears in the corner of a website. The public-resource and failure paths are the parts that decide whether it is actually finished.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.