> For the complete documentation index, see [llms.txt](https://docs.glbnxt.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.glbnxt.com/workspace/agents-and-assistants/glbnxt-agents.md).

# GLBNXT agents

GLBNXT agents are ready-to-use AI agents in your GLBNXT chat app. Choose an agent, ask your question and get to work. There is nothing to configure, no prompts to engineer and no tools to connect.

## Why three agents instead of one

A quick lookup and the analysis of a 600-page tender are different jobs. One agent that tries to do both is either too slow for the first or not thorough enough for the second. So we built three agents, each tuned for a type of work:

* **Lunar** for fast answers to quick questions
* **Nexus** for everyday work with documents and research
* **Atlas** for complex work that needs the most capable models

You pick the agent that fits the task, and switch whenever the task changes.

## Everything an agent needs, already built in

Building a useful AI agent takes more than a language model. It needs to read files, understand images, handle voice, search the web, research in depth and produce real output. We prepared all of this for you, so every GLBNXT agent works out of the box:

* **Multimodal input:** documents, spreadsheets, presentations, images and scans
* **Voice and audio:** recordings are transcribed automatically
* **Web search with sources:** every fact is numbered and linked to where it came from
* **Deep Thinking:** in-depth research for questions with many parts
* **Real output:** Word, PDF, PowerPoint and Excel files, charts, tables and images
* **Exact calculations:** numbers and dates are calculated, not estimated

## Hosted in the EU, kept up to date by GLBNXT

{% hint style="info" %}
**Always hosted in the EU.** GLBNXT only assigns AI models that are hosted in the European Union. This applies to every agent, now and after every upgrade.
{% endhint %}

You choose an agent, not a model. GLBNXT selects the AI models behind each agent and keeps them current, so you always work with the most suitable models for the task. See [The agent stays, the model improves](#the-agent-stays-the-model-improves).

***

## Compare the agents

The table shows where the agents differ. Everything they have in common is listed below it.

|                                                         | [Lunar](https://docs.glbnxt.com/workspace/agents-and-assistants/glbnxt-agents/lunar.md) | [Nexus](https://docs.glbnxt.com/workspace/agents-and-assistants/glbnxt-agents/nexus.md) | [Atlas](https://docs.glbnxt.com/workspace/agents-and-assistants/glbnxt-agents/atlas.md) |
| ------------------------------------------------------- | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- |
| **Performance**                                         |                                                                                         |                                                                                         |                                                                                         |
| Depth²                                                  | 70                                                                                      | 100                                                                                     | 140                                                                                     |
| Speed³                                                  | 300                                                                                     | 100                                                                                     | 60                                                                                      |
| In short                                                | **Fastest**                                                                             | **Balanced**                                                                            | **Deepest**                                                                             |
| **Hosting and jurisdiction**                            |                                                                                         |                                                                                         |                                                                                         |
| Hosted in                                               | EU                                                                                      | EU                                                                                      | EU                                                                                      |
| Model providers                                         | EU-based                                                                                | EU-based                                                                                | EU and non-EU¹                                                                          |
| Jurisdiction                                            | EU only                                                                                 | EU only                                                                                 | EU & non-EU¹                                                                            |
| **Answers and research**                                |                                                                                         |                                                                                         |                                                                                         |
| Answer style                                            | 1–3 sentences, often within a second                                                    | Clear answers with sources                                                              | Thorough answers with sources                                                           |
| Web sources per question                                | Up to 4                                                                                 | Up to 6                                                                                 | Up to 8                                                                                 |
| Deep Thinking                                           | Compact                                                                                 | Parallel research                                                                       | Most parallel research                                                                  |
| Document size                                           | 25 MB or 500 pages                                                                      | 25 MB or 500 pages                                                                      | 25 MB or 1,000 pages                                                                    |
| **Extras**                                              |                                                                                         |                                                                                         |                                                                                         |
| Download chart data as Excel                            | <i class="fa-xmark">:xmark:</i>                                                         | <i class="fa-check">:check:</i>                                                         | <i class="fa-check">:check:</i>                                                         |
| Ask a follow-up question when your request is ambiguous | <i class="fa-xmark">:xmark:</i>                                                         | <i class="fa-check">:check:</i>                                                         | <i class="fa-check">:check:</i>                                                         |
| Images                                                  | Create                                                                                  | Create                                                                                  | Create and edit                                                                         |

¹ Atlas can include models from providers that fall under a non-EU legal regime. These models are always hosted in the EU.

² **Depth** is the quality and completeness of an agent's answers on a fixed GLBNXT test set of realistic tasks in Dutch and English: answering questions from documents with correct citations, researching questions with several parts, calculating exactly and producing usable files. Nexus is the reference at 100. A score of 140 means 40% more depth than Nexus.

³ **Speed** is the median time an agent takes to answer the same test set, indexed so a higher score means a faster answer. Nexus is the reference at 100. A score of 300 means about three times as fast as Nexus.

Depth and Speed are indicative. We measure the agent as a whole, including its tools, and measure again after every upgrade.

### Every agent can

* Read PDF, Word, Excel and PowerPoint files (including older Office formats), images and scans
* Transcribe voice and audio recordings up to 100 MB (roughly 2–3 hours)
* Search the web and cite every source
* Make Word, PDF, PowerPoint and Excel files
* Make charts, tables and images (1024 × 1024)
* Calculate exactly, including dates
* Work in Deep Thinking mode for questions with many parts

***

## Which agent should I use?

* **Use Lunar** for quick questions and fast answers. It is about three times as fast as Nexus.
* **Use Nexus** for most work. It's the everyday choice: hosted and provided within the EU, under EU jurisdiction only.
* **Use Atlas** for complex work, such as long documents and thorough analysis. It goes about 40% deeper than Nexus.

If your request suits another agent better, the agent tells you which one to use, and whether switching changes the jurisdiction that applies.

***

## The agent stays, the model improves

You choose an agent, not a model. Each agent has a fixed role and a fixed set of capabilities. Behind it, GLBNXT always assigns the AI models that are most suitable for that role and those capabilities.

AI models improve every few months. We test new models continuously, and when a model does the job better, we move the agent onto it. You don't retrain your team, change your workflows or redo your assessments. The agent simply gets better.

When we upgrade an agent:

* **Hosting stays in the EU.** We only assign models hosted in the European Union.
* **The jurisdiction stays the same.** An upgrade never changes an agent's jurisdiction.
* **Capabilities stay or grow.** Tools, limits and file types never shrink.
* **Depth and Speed are measured again,** so the scores in this documentation stay accurate.

***

## Start a chat with an agent

1. Open your organisation's GLBNXT chat app.
2. Open the model picker and choose an agent: **Lunar**, **Nexus** or **Atlas**.
3. Type or speak your question. To work with files, attach them to your message.
4. While the agent works, status lines show what it is doing, for example "Searching the web…" or "Making the Word document…".
5. Read the answer. It lists its sources and, when the agent made a file, gives you a download link.

***

## Sources you can check

When an agent uses a document or web page, it marks the fact with a number, such as \[1] or \[2], and lists the sources under the answer. For documents, it gives the page number. For web pages, it only cites pages it actually opened, so every link works.

***

## Files the agent makes for you

Ask for a Word document, PDF, PowerPoint presentation or Excel sheet and the agent makes it and gives you a download link. Download the file soon: files are removed after 30 days.

***

## Deep Thinking

Some questions need real research, for example comparing several options or answering a question with many parts. For these, choose the **Deep Thinking** version of an agent. Every agent has one, listed in the model picker with **· Deep Thinking** after its name, for example **Nexus · Deep Thinking**.

Deep Thinking splits your question into parts, researches them in parallel, checks its own draft and fills any gaps before it answers. This takes a few minutes. You see every step while it works.

How far Deep Thinking goes depends on the agent:

* **Lunar:** compact. It researches fewer parts at the same time and finishes sooner.
* **Nexus:** parallel research across several parts of your question.
* **Atlas:** the most parallel research, for the broadest questions.

For a complex question, Lunar suggests its Deep Thinking version.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.glbnxt.com/workspace/agents-and-assistants/glbnxt-agents.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
