Your AI Glossary

Agent

AI programs that independently perform tasks or make decisions.

Assistants

Digital helpers that take over recurring tasks, create content, or answer questions.

We already provide several assistants for you in the KI-Workplace. In addition, your admins — and you yourself — can create your own. Select an assistant via the “Assistants” button at the top of the chat view, or create your own via the Assistant & Model settings. You can find more information in the Assistants Guide.

• Global Assistants Assistants we provide for all users.
• Company Assistants Assistants available to all employees in your company. They can only be created or modified by admins.
• Personal Assistants Assistants you create for yourself and that only you can see.

Create a copy: In the Assistant & Model settings, you can create your own copy of any assistant (under Actions).

Admins can also change an assistant’s type between Company and Personal.

Context Display

A percentage indicator above the chat input field showing how much of the Context Window is already in use. As it approaches 100 percent, it’s advisable to start a new chat.

Context Window

The model’s “working memory.” The KI keeps all prompts and files of a conversation in the Context Window. When it is full, the chat ends. For this reason, it’s recommended to finish chats before the Context Display reaches 100%.

Hallucination

When the AI invents incorrect information that appears plausible. Any AI can produce hallucinations, which is why we always recommend reviewing the output. They can be reduced with precise prompts and suitable knowledge sources.

Instance

A dedicated, isolated AI environment. Each customer receives a separate, protected workspace that no one else can access.

Knowledge

A pool of files used as a knowledge source for an assistant. You can attach Knowledge to an assistant so the AI can look up relevant information when answering questions. It functions like retrievable knowledge that is not fully loaded into the conversation but accessed in parts depending on relevance. The underlying technology is called RAG (see below).

LLM (Large Language Model)

A language model that understands and generates text. It is the core of every chat interaction. Speed, quality, and context length vary depending on the model.

MCP (Model Context Protocol)

A standard that allows AI models to securely retrieve data from various systems. It serves as the technical foundation for integrating external tools such as files, emails, databases, and more.

Models

Individually trained AI systems with their own strengths. The KI-Workplace provides several current open-source models, with the most powerful one enabled by default.

Multimodal Models

Models that can process text, images, audio, or video simultaneously.

n8n

A workflow automation tool. It can be connected to AI models via MCP and will soon be accessible through the KI-Workplace to trigger complex processes. n8n is also the name of the company behind the tool, based in Berlin.

Open-Source Models

AI models with open code, free to use and modify. The KI-Workplace exclusively uses transparent open-source models, without dependence on US providers.

Prompt

The input or task you give to the AI.

RAG (Retrieval Augmented Generation)

A technology that enables the AI to access external knowledge. In the KI-Workplace, you can store files as Knowledge in every assistant. This is especially useful when the knowledge is extensive and the AI should only retrieve the parts that are relevant at a given moment.

Reasoning

The AI’s ability to think logically and draw step-by-step conclusions.

Data Sovereignty

The assurance that your data, chats, and assistants are protected and only you control them. This is a top priority in the KI-Workplace. No sensitive data leaves your company; everything is processed locally and in compliance with the GDPR — ideal even for personal and confidential data and files.

System Prompt

Baseline instructions that define how the AI should behave (e.g., role, goal, and rules). Models have their own system prompts that remain in the background and generally instruct them to be helpful.

Each assistant also receives its own system prompt (a field in the creation form). It explains the assistant’s task and can provide specific knowledge. During testing, you should refine the system prompt until the assistant fulfills its purpose optimally.

Tool Calling

The AI can call external tools (e.g., web search, calendars, databases). This enables intelligent automation, allowing assistants to independently perform actions or retrieve data.

Token

The smallest processing unit of a model (text, image, or audio fragment). All chat inputs and uploaded files are converted into tokens so the AI can use them.

A rule of thumb: one token is roughly four characters; 100 tokens are about 75 words; one A4 page is about 375 tokens. However, actual counts may vary widely—especially for files, where complexity matters more than size.

Training

The process by which a model learns from large amounts of data. The models in the KI-Workplace are fully pre-trained—no customer data is used for further training.

Web Search

The AI searches the internet in real time for up-to-date information. You can activate it in the KI-Workplace via the “Web search” button in the chat window. Because parts of the conversation may be sent to the search service, you must explicitly enable or allow web search each time.