AGENT PROFILES

DeutschlandGPT

DeutschlandGPT GmbH · Agent builders · Agent workspaces · Coding agents · Agent orchestration

German AI platform with chat, agent mode, Cowork and Code on the desktop as well as workflows, hosted according to the provider in the Open Telekom Cloud.

Typical tasks

Availability in Europe
Added to the catalog on German AI platform with chat, agent mode, Cowork and Code on the desktop as well as workflows, hosted according to the provider in the Open Telekom Cloud.Details
01

What it does

7 features

DeutschlandGPT · AI work platform with agent mode, desktop agents and workflows

Product and access

DeutschlandGPT is a proprietary product of DeutschlandGPT GmbH, which according to the trust page belongs to Titanom Group GmbH. The platform brings models from several providers into one interface.

Core features

7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.

  • Agent mode in chat

    For each message you set whether the AI only reads, researches the web or acts in your systems, such as sending emails or creating Jira tickets. Acting can be controlled per integration.

    Source 5
  • Cowork in local folders

    You name a goal, Cowork plans the steps, shows the plan and works in shared folders. It creates Excel, Word, PDF and PowerPoint files. It runs as a desktop app for macOS, Windows and Linux.

    Source 6
  • Code for software tasks

    The coding agent searches a project, changes files, runs builds, tests and Git, and checks its result in a preview. You see every step.

    Source 7
  • Workflows with approval

    Triggers are manual, scheduled, webhook, form, integration or subworkflow. An approval switch on a node holds the run until a person decides.

    Source 17
  • Special applications

    Custom assistants with behaviour, knowledge and tools that can also act according to the page. There are four sharing levels from private to editable across the organisation. The page's address calls them "agents".

    Source 3
  • Integrations

    The page lists 87 tools and data sources, the FAQ "over 80". The modes Search (read only) and Actions (read and write) can be released separately per tool.

    Source 8
Show all 7 features
  • Roles and permissions

    According to the page, rights run through roles with over 90 individual permissions, with SAML SSO and SCIM from Business. Security-relevant actions are logged and can be exported.

    Source 9

Usable AI models

Fixed choice across providers

You pick from a dropdown of around 50 language models from OpenAI, Anthropic, Google and Mistral, plus open models in the Open Telekom Cloud. The free plan includes only selected models, and the choice shrinks if your organization activates the §203 professional-secrecy feature. The pages checked do not mention your own keys or models.

Model families named

  • OpenAI (GPT)
  • Anthropic (Claude)
  • Google (Gemini)
  • Mistral
  • Alibaba (Qwen)
  • Nvidia (Nemotron)
  • Z.ai (GLM)
Model choice
You choose
Your own API key
Not stated
Router or custom endpoint
Not stated
Local models
Not stated
Depends on
Plan, Admin approval

The rating covers the generally available main access. Documented · Sources accessed on

DeutschlandGPT: Modelle · DeutschlandGPT: FAQ · DeutschlandGPT: Preise · DeutschlandGPT Docs: Organisation

02

Does it fit my task?

2 practical tasks · 6 requirements and limits

A good fit if …

  • you want to use agent work on the desktop, in chat and in workflows from one account
  • you want to check sub-processors and model routes on public pages before signing
  • you want to start with a free account without a sales call

Less suitable if …

  • you have to run the platform in your own data centre
  • you require a SOC 2 attestation or a completed security questionnaire from the provider
  • you need documented retries and error paths for workflows

Editorial assessment based on provider documentation. Not a product test.

Practical tasks

Editorial examples based on documented features. No product tests performed.

01

Create a report from a folder

Your input
A shared folder with source files and a goal instead of individual steps.
Possible result
Cowork shows the plan and then creates a file in the folder.
What to check
Confirm each change in Normal mode and check the result in the change view.
02

Ticket intake with approval

Your input
A workflow with an integration trigger, an AI step and an approval switch.
Possible result
The run holds at the node until a person approves or rejects.
What to check
Check status, duration and cost per node in the history and set a spending limit.
EDITORIAL EXAMPLE

Have Cowork create an Excel report with working formulas from a shared folder.

Your possible prompt

Compare three providers for [task]. Use their official documentation. Show differences, sources, and open questions in a table.

Adjust the task, data and permissions.

Requirements and limits

  • Limitation

    Auto mode without asking

    In Auto mode actions run without asking. A review model checks risky commands, but deleting and overwriting need no confirmation. Completed actions are not reverted when you stop a run.

    Source 20
  • Limitation

    No self-operation

    The trust page rules out operation in your own data centre. Cowork and Code run locally on the device, the rest of the platform with the provider.

    Source 10
  • Requirement

    Access to Cowork and Code through roles

    Role permissions control access to Cowork and Code. Commands are available in Cowork only if your account may also use Code.

    Source 19
  • Limitation

    Budget per plan

    Each plan includes a monthly AI budget. Multi-step runs use more. After that you switch to pay-as-you-go or, on Business and Enterprise, to cheaper models.

    Source 12
  • Not yet confirmed

    Audit log by plan

    The governance page describes exportable logs, the pricing page lists the audit log under Enterprise. It is not clear from this what Business includes.

    Source 2
  • Not yet confirmed

    Error paths in workflows

    Failed nodes show their error in the history. The workflow documentation does not describe retries or error paths.

    Source 17
03

Can I use it?

With requirements

Europe & regional availability

Trust page, data processing annex, models page and FAQ reviewed. All statements come from the provider; certificates and contracts with model providers were not seen. The trust page dates one agreement after its own revision date.

Reviewed scope: Platform with the provider in Germany; models through EU access
Reviewed on:

  • Available

    Storage in Germany

    According to the trust page history, files and settings are stored in the Open Telekom Cloud in Magdeburg and Biere, operated by T-Systems. The data processing terms state persistent data only in German data centres. According to those terms administrative access and support come from the EU.

    Region, reason and source

    Region covered: Germany

    Documented reason: Provider statement in the trust page and data processing terms. The BSI C5 attestation belongs to T-Systems, not DeutschlandGPT.

    Official source · deutschlandgpt.de · Official source · deutschlandgpt.de

  • With requirements

    Model processing in the EU

    According to the trust page models run through EU access: OpenAI through Azure and OpenAI Ireland, Claude through AWS, Gemini through Google Vertex, plus Mistral. Open models run in the Open Telekom Cloud. For all model providers the provider cites agreed zero data retention and no training.

    Region, reason and source

    Region covered: EU

    Documented reason: Contracts with model providers cannot be viewed. For US providers processing runs through EU subsidiaries according to the page.

    Official source · deutschlandgpt.de · Official source · deutschlandgpt.de · Official source · deutschlandgpt.de

  • Available

    Data processing agreement and sub-processors

    The data processing agreement under Art. 28 GDPR is publicly readable. Annex IV lists T-Systems, Hetzner, IONOS, Microsoft Ireland, AWS EMEA, Google Cloud EMEA, Mistral AI, OpenAI Ireland, Linkup and Sentry.

    Region, reason and source

    Region covered: EU/EEA

    Documented reason: The contract text itself documents the list.

    Official source · deutschlandgpt.de

  • With requirements

    "The data never leaves the EU"

    The FAQ and trust page say so. The data processing terms name as an exception the error monitoring by Sentry, based in the USA, without user content and in an EU data centre, based on standard contractual clauses.

    Region, reason and source

    Region covered: EU and USA

    Documented reason: "Outside US jurisdiction" is a provider claim. The pages give no legal assessment.

    Official source · deutschlandgpt.de · Official source · deutschlandgpt.de · Official source · deutschlandgpt.de · Official source · deutschlandgpt.de

  • Availability still unclear

    Certificates and limits

    According to the trust page ISO/IEC 27001:2022 and ISO 9001:2015 from TÜV SÜD are issued to the parent company Titanom Group GmbH, with DeutschlandGPT GmbH listed as a site. The page names no SOC 2 report. "GDPR compliant" remains a provider label.

    Region, reason and source

    Region covered: EU/EEA

    Documented reason: Certificates not viewed ourselves; no authority assessment found.

    Official source · deutschlandgpt.de

Europe and regulation explained

Accounts & permissions

Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.

Configuration covered: Hosted platform with role management; Cowork and Code on the device

Sources reviewed:

Show all 3 points
  • Documented

    Account used

    SAML SSO and SCIM are available from the Business plan. Organisations can require two-factor authentication. According to the documentation the desktop app stores tokens in the operating system's credential store.

    Evidence · Evidence · Evidence
  • Documented

    Admin approval

    Administrators enable integrations across the organisation and release Search and Actions separately per tool. Chat defaults govern agent mode, tool policies govern new integrations and local MCP servers.

    Evidence · Evidence
  • Documented with conditions

    Data access & team sharing

    According to the page integrations take over the access rights of the connected system. Cowork and Code work only in shared folders; protected locations such as .ssh or .aws stay blocked even there. According to the documentation the command check does not detect paths a program builds at runtime.

    Evidence · Evidence
04

Memory and conversation continuity

2 of 6 criteria documented

According to the documentation the AI remembers only two profile fields from one conversation to the next. Projects and folders provide lasting context.

Two profile fields and context per project; no dedicated memory feature documented

Sources accessed:

Documentation by criterion

StorageDocumented

Storage and scope

Two free-text fields under Settings, Profile, each up to 2,048 characters. According to the trust page the data is stored in the Open Telekom Cloud in Germany.

SelectionDocumented

Selection and timing

You fill in the fields yourself and save them. The documentation does not describe automatic remembering.

Edit & deletePartly documented

Inspect, edit and delete

You edit or clear the fields in your profile. The page describes no admin view of them.

Access boundariesPartly documented

Agent and channel boundaries

The fields feed into every conversation. According to the FAQ the context of a client matter persists across its chats but does not move between matters. The documentation does not say whether the fields also apply in Cowork and Code.

RetrievalPartly documented

Retrieval

According to the documentation the profile fields are included in every conversation. The reviewed page does not say how project context is selected.

New conversationsPartly documented

Session reset and compaction

The reviewed pages say nothing about compaction. Subagent transcripts survive restarts and are deleted after 30 days or with the conversation.

05

Cost and effort

4 plans · Prices as of 5 November 2026

Prices and included usage

Pricing checked

Prices per user plus VAT. Free stays free of charge. Workflows, API and chat widget cost extra by usage.

PlanPriceIncluded
FreeEUR 0

Limited usage, special applications and projects. Cowork, Code and integrations are included, but not meeting transcription, user administration and add-ins.

BusinessEUR 24 per user per month, EUR 20 annually

Cowork, Code, meeting transcription, user administration, SAML SSO, SCIM and configurable deletion periods. Professional secrecy mode at no extra charge.

Business 5×The page lists EUR 100; it does not clearly state the billing period there

Five times higher limits.

EnterpriseCustom on request

Everything in Business, plus dedicated onboarding, white labelling, payment by invoice, personal consulting, audit log and chat widget.

Price source

  • The FAQ says "from EUR 19", the plan page EUR 20 and EUR 24. Some pages mention a "Pro" plan that the plan overview does not list.
  • The integrations page says all integrations are in every plan, but only one can be connected on the free plan.
  • Third-party price information was not used.

Setup

Create a free account and start in chat or the desktop app. For actions you connect integrations, usually by OAuth; administrators set rules and roles.

Operation & oversight

Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.

Deployment covered: Platform with the provider; Cowork and Code as a desktop app on your device

Sources reviewed:

Editorial assessment

DeutschlandGPT discloses sub-processors, model routes and the rights of desktop agents and names concrete limits. Error paths in workflows are not documented. You should check the promise "outside US jurisdiction" against the data processing terms, because they also list EU companies of US providers and Sentry, based in the USA.

Show all 6 points
  • Documented

    Files and working state

    Cowork works in a local folder that persists between sessions. Change history is kept for 30 days or up to 1 GB. For conversations you choose a period or "never"; according to the solutions page the platform does not delete automatically by default.

    Evidence · Evidence · Evidence
  • Documented

    Sign-ins

    The desktop app signs in through the browser. In the organisation SAML SSO, SCIM and optionally enforced two-factor authentication apply. Most integrations use OAuth.

    Evidence · Evidence · Evidence · Evidence
  • Documented

    Starting further runs

    Workflows start manually, on a schedule, by webhook, form, integration or as a subworkflow once a version is deployed. For Cowork the product page names no schedules and points to Workflows. A "Don't ask" mode is intended for unattended runs.

    Evidence · Evidence · Evidence
  • Documented

    Human intervention

    In Normal mode Cowork asks before changes and commands, in Plan mode it changes nothing, in Auto mode it does not ask. In workflows an approval switch holds the run. In chat you control acting per integration.

    Evidence · Evidence · Evidence
  • Documented

    Detecting errors

    The workflow history shows status, duration, cost and the error per node. The desktop app marks sessions as waiting, running, failed or done. The page says nothing about how long workflow logs are kept.

    Evidence · Evidence
  • Documented with conditions

    Resuming after failures

    If a Cowork run reaches its step budget, "Weiterarbeiten" (continue working) resumes the same run. Stopping ends it, and completed actions stay. For workflows the documentation describes no retry and no error path.

    Evidence · Evidence
06

Evidence and open questions

23 sources · 3 open points

Open points

The reviewed sources contain no reliable information on these points. A gap says nothing about the product.

Europe: Availability still unclear

Requirements and limits: not yet confirmed

Sources and evidence

Official website, pricing page, trust page, data processing terms and documentation reviewed. Feature descriptions and security statements come from the provider, and we did not view the certificates ourselves; example tasks are editorial. Our own hands-on tests are still pending.

Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.

Read profile as Markdown

How we assess the sources

Provider

DeutschlandGPT GmbH offers an AI platform for organisations. According to the trust page its parent company is Titanom Group GmbH.

Operator
DeutschlandGPT GmbH
Organization
Provider
Provider sourcesSource 1 Source 2 Reviewed
Changes to this profile 1
  • German AI platform with chat, agent mode, Cowork and Code on the desktop as well as workflows, hosted according to the provider in the Open Telekom Cloud.

These notes describe editorial changes. Source and pricing review dates appear in their respective sections.