Features

One AI for every task in your organisation.

You put in text, documents, speech or images and get back answers with sources, drafts and finished documents.

Question“Which deadlines apply?”
DocumentPhysician_letter.pdf
SpeechMeeting recording
Image or scanPhoto of a form

baseboxChat · Apps · Workflows

Answer with sourcewith the passage
Summaryon one page
Transcriptfrom the recording
Finished documentas a Word file

All features are included in every licence.

Overview in 1 minute

Everything at a glance.

Apps

Control

Sign-in via your directory
Permissions from your structure
Approval by people

Connection

Connectors included
Read-only by default
With each person's permissions

AI model

Selected and tested
Sends nothing outside
Switch by update

Apps

What your staff do with it

Chat assistantAsk, write, translate, summarise.

Answers from your documentsWith a source for every statement. Sources →

Text recognitionReads scanned documents and images.

Speech recognitionConverts speech to text.

Understand images and filesUpload and analyse PDFs, spreadsheets, images and audio.

Create documentsDrafts straight as a file.

Knowledge basesCollections of your documents that apps and chats access.

Web searchCurrent information from the web, if your administrator enables it. Security of web search →

Thinking modeQuick answer or thorough thinking; you choose for each question.

An app for every task

Your own appsRecurring tasks as an app, without programming, in a few minutes.

App StoreTake ready-made apps from a library and publish them in your organisation. Docs ↗

Hundreds of use casesFrom everyday hospital and administrative work. Hospital → Administration →

Groups and sharingShare apps and knowledge with teams.

Interface for developersOpenAI-compatible, with a separate key per person. For research →

App Builder

You describe a task, and AI agents turn it into an app that handles it from then on, even across several steps and systems. The agents work only within your control. From November 2026.

How an app comes about: build, share, use, extend

Your staff can only use what your administrator has enabled: apps, connections to systems and AI models. This also applies to the App Builder.

App

Instruction
Documents
Connections

or from the App Store

Groups

Groupuse
Groupuse
Personedit

People

Person A
Person B
Person C

own access, own data only

Workflow

Step 1
Step 2
Step 3

only what is enabled

Build

An app consists of an instruction, documents and connections to your systems. You build it without programming, or you take a ready-made app from the App Store.

Share

The author shares the app with groups from your user directory or with individual people. The author decides whether they may only use the app or also edit it, i.e. change the instruction, documents and connections.

Use

When the app accesses a system, each person signs in there with their own access. They see only what they are allowed to see in that system anyway. Sharing an app shares its documents, never your own access.

Extend

In the App Builder you describe a task, and AI agents turn it into an app, even across several steps. The agents pick the right connections themselves, but only use what your administrator has enabled. From November 2026.

And the AI models?

Models work like connections to systems: your administrator enables them, and then they can be selected in an app. Usually there is one. If you want a specialised model, for example one trained on your tasks, we set it up and your administrator enables it.

All questions on permissions →

Control

The more AI does on its own, the more important it is who may do what. basebox checks every action before it happens, for chats, apps and agents alike.

Sign-in via your user directoryActive Directory, LDAP, SAML or OIDC, optionally with a second factor.

Permissions from your structureEveryone sees only what they are allowed to see anyway. Identity and permissions in detail →

Agents with the person's permissionsNo agent may do more than the person using it.

Approval by a humanWrite actions by agents run only once a person confirms them.

Audit logSign-ins, changes to accounts, permissions and apps, exports and connector calls. Entries cannot be deleted individually; export as CSV. What is logged →

Audit log for your SIEMExport on request for Splunk, Elastic, Sentinel or QRadar, not yet automatic. Docs ↗

Models under controlExternal models only if your administrator enables them, with processing location. Data flow →

Rules for the whole organisationUsage policies and system defaults, set centrally.

Overview for administratorsDashboard, daily reports and data export.

Connection

All connectors from our library are included in your licence, including new ones. Switch on, sign in with your own access, use.

OrbisF1
roXtra
FHIR
SharePoint
Exchange
Nextcloud
Network drives
Confluence
Jira
YouTrack
DokuWiki
CentralStationCRM
Web search
Create documents
Calculator

All connectors by area, with prices →

If a system is missing, we build the connectorDesign free of charge, then a fixed price. If we build it new and it may go into our library, you pay half.

Read-only by defaultWrite access only after agreement.

With each person's permissionsNo shared service account unless you want one. Every call is in the audit log.

Open standardThe connectors work with MCP.

AI model

We choose the model, you get good answers.

Selected and testedWe test new models with tasks from hospitals and administration and deploy the best one.

Sends nothing outsideThe AI model in basebox runs on our servers or yours, without a connection to the internet. By default, we do not send your data to any AI model of another provider.

Switch by updateA new model arrives by update; apps, permissions and data stay the same. Your data does not train any model.

New models arrive by update, tested by us beforehand.

basebox GmbH tests

Model Bnewly released

Chat
Long texts
Document search
Tool calls
German
Load

Update, even without internet →

Your protected area

basebox

AppsControlConnection

AI models

Model ADefault

Model Binstalled

What platform, model and computing power each provide →

Frequently asked questions

Which AI models does basebox use?

basebox uses open-source models that we run ourselves: in the cloud on our servers in Germany, with the server on your hardware. We do not develop our own model. We install the best, most capable models that suit the available hardware, including on our own; there are no feature tiers. On request, we also set up an app that works with an external model, for example from Google or Anthropic, processed in Frankfurt. Your organisation can also connect interfaces of other providers. What this means for data protection →

How quickly do new models arrive, for example from OpenAI or Anthropic?

We replace the AI model in basebox when a clearly better one appears, after our tests with tasks from hospitals and administration. That does not happen all the time. We update the external model in its own app within a few days once a new one is available and tested. For this we use Google Vertex in Frankfurt: no training with your inputs, logging by Google for abuse detection for at most 30 days. With other providers, such as STACKIT, inputs are not stored at all. If you want to connect other providers such as OpenAI, that is possible. We then point out the consequences for data protection and security.

Can all models research on the internet?

Yes. Web search is a feature of basebox, not of the model. If your administrator enables it, it works with any model. Web search never sees your internal data. Security of web search →

Can we use basebox in hybrid mode, with external models for individual groups?

Yes. For sensitive data, everyone works with the AI model in basebox. In addition, we set up apps with an external model, for example from Anthropic, OpenAI or Google. Your administrator decides which people and groups use such an app and which documents and connectors it gets. This works in the cloud and on the server, provided your firewall allows the connection to the provider. External models are billed by usage; if you include usage in your licence, the costs are covered within the agreed scope.

Can our developers program with basebox?

Yes, on the basebox server. Your developers keep their tools in VS Code or in the terminal, such as Cline, Kilo Code or Aider, and connect them via the OpenAI-compatible interface of basebox. A coding model then runs on your GPUs alongside the AI model for the whole organisation; the code does not leave the building. For this, the server needs room for that model; we work out with you how many developers can work at the same time. In the cloud, the quota is enough for questions about code, not for agents that program autonomously. What works and what does not →

All questions →

Last updated: 29 October 2026