> ## 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.

# Knowledge ai

AI Chatbot reuses RAUM's organization-level AI and knowledge infrastructure, but only the **publicly eligible subset** reaches visitors.

## Procedures

Chatbot can use approved Procedures for:

* public policy;
* factual support rules;
* troubleshooting;
* decision logic;
* escalation conditions; and
* approved public-tool workflows.

A Procedure must pass Chatbot publication review before the public pipeline can retrieve it.

For sensitive procedures, public policy can additionally require verified identity, visitor confirmation, specific widgets, or public-action eligibility.

[Shared Knowledge →](https://docs.raum.am/chatbot/shared-knowledge.md)

## AI Guidelines

Chatbot uses approved **Response** AI Guidelines for communication behavior.

Use them for:

* public tone;
* terminology;
* response structure;
* wording constraints;
* domain/widget-specific behavior; and
* safety or disclosure instructions.

Assistant **Translate** use-case guidelines are not Chatbot translation rules.

## Documents & Websites

Uploaded documents and indexed websites can become Chatbot knowledge after public approval.

Use them for customer-safe reference material that does not need the explicit branching/actions of a Procedure.

Do not approve an internal document merely because most of it is harmless. Public publication is resource-level trust, not a scavenger hunt where the model is expected to avoid the one confidential paragraph.

## Gallery

Approved Gallery assets can support public answers with files or images where the widget and resource policy allow them.

Use public Gallery content for customer-safe assets such as:

* screenshots;
* forms;
* diagrams;
* PDFs; or
* links/media associated with a support procedure.

Assistant-specific Zendesk targeting is not itself a public Chatbot permission rule.

## Updates

Temporary guidance can be approved for Chatbot when customers should receive it.

Examples:

* incident notice;
* temporary shipping delay;
* short-term product limitation; or
* time-bound public policy exception.

Resolve/expire the Update when it is no longer valid so temporary public guidance does not become accidental doctrine.

## AI models

Chatbot uses the generation and embedding configurations under **AI → Models**.

Generation models write answers and reason over the request. Embedding models power semantic retrieval.

Provider credentials belong to your organization and are not exposed in the widget.

## Connections

Internal API Connections and MCP Servers remain private by default.

A connection can become available to website chat only when it satisfies the separate public connection policy and Chatbot publication review.

Public SOURCE access can be restricted by widget, identity tier, bounded input rules, and bounded output projection.

Public WRITE actions use a stricter policy and require host-signed identity plus customer confirmation, idempotency, bounded inputs, rate limits, timeouts, and audit.

[Public connections & actions →](https://docs.raum.am/chatbot/public-connections.md)

## Features that do not carry over

The following Assistant features are not part of the public Chatbot reasoning context:

* Brand Voice;
* AI Coaching;
* Zendesk Ticket Fields;
* Zendesk Macros;
* Zendesk groups/organizations;
* Assistant Translate-only guidelines.

This separation is deliberate. A website visitor and a Zendesk agent are not the same trust boundary, despite both being excellent at finding edge cases five minutes after launch.


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