What it does
7 features
innoGPT · AI work platform with assistants, tasks and workflows
Product and access
innoGPT is a proprietary product of Inno KI GmbH in Vechta. The platform combines models from several providers. The data processing article and the security page spell the company "innoki GmbH".
Core features
7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.
Assistants with tools
An assistant combines a role, files and tools such as web search, Code Interpreter, Canvas, connected apps and MCP servers. Skills are automatically active in every chat with it.
Source 11Subagents
An assistant can delegate subtasks to other assistants. You address them with @, or the main assistant loads them automatically.
Source 11Custom Actions
An API can be connected through an OpenAPI, YAML or JSON specification, including for write calls. According to the page the agent chooses the endpoint and parameters itself. Only admins can set this up, and it is not available in normal chat.
Source 6Scheduled tasks
According to the home page recurring processes such as reports or data maintenance start by schedule or event. According to the changelog routines always ask for approval before actions outside innoGPT.
Source 1Workflows in public beta
According to the changelog workflows have been available as a public beta in every workspace since 14 October 2026 and support native triggers from Microsoft 365. The automation page still marks "Workflows & API" as "Coming soon".
Source 12Integrations and MCP
The page lists over 1,000 tools and native MCP support for your own servers. Local MCP servers in the desktop app can be allowed, tied to confirmation or blocked per tool.
Source 6
Show all 7 features
Roles and model access
Default roles are Admin, Member and Guest, plus custom roles. Model access can be limited per role, for example to models in the EU. According to the documentation SCIM is still in progress.
Source 16
Usable AI models
Fixed choice across providers
You choose from more than 50 models or let Smart Select decide. Only innoGPT Standard is unlimited; it bundles open models in the EU without naming them. Premium models draw on your plan's budget, and admins set which models are available. The pages mention no keys of your own, and it remains open whether Enterprise self-hosting allows your own models.
Model families named
- OpenAI (GPT)
- Anthropic (Claude)
- Google (Gemini)
- Mistral
- xAI (Grok)
- DeepSeek
- Meta (Llama, Muse)
- Cohere
- Moonshot (Kimi)
- Z.ai (GLM)
- Model choice
- Automatic or your choice
- 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
innoGPT: AI models · innoGPT: Core · innoGPT: Preis · innoGPT Docs: Sprach- und Bildmodelle · innoGPT Docs: innoGPT Standard
Does it fit my task?
2 practical tasks · 6 requirements and limits
A good fit if …
- you want to build assistants with tools, MCP servers and subagents in one workspace
- admins should limit model choice per role to zones such as the EU
- you want scheduled tasks with approval before external actions
Less suitable if …
- you need to use workflows in production without a beta caveat
- you want to read the data processing agreement, technical measures and full sub-processor list without an NDA or request
- you require SCIM for user management
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Create a ticket through an assistant
- Your input
- An assistant with a Custom Action on the ticket system and a clear role.
- Possible result
- The assistant chooses the endpoint and parameters and creates the ticket.
- What to check
- Check each action in the audit log and keep role-based approvals narrow.
Recurring report
- Your input
- A scheduled task with a fixed brief, for example a monthly report.
- Possible result
- The task runs on schedule and delivers the result as its own run.
- What to check
- Watch for approvals before external actions and check the chosen model's hosting zone.
Have an assistant with a Custom Action create a ticket from a request.
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
Workflows still in beta
According to the changelog workflows are a public beta. The terms provide beta features "as is", without a guarantee of being error-free.
Source 9Limitation
Not all models in Europe
The home page says the software and "(most)" AI models are hosted in Europe. Admins choose zones such as EU/EU, EU/Global or Global/Global. According to the terms, choosing models outside the EU counts as the customer's consent to the transfer.
Source 1Requirement
SSO costs extra on Business
The pricing page lists SSO/SAML on the Business plan as an extra charge, without an amount.
Source 2Limitation
No agent details in the plans
The pricing page mentions agents and automations in no plan. Business promises "all features", and it names no run limits.
Source 2Not yet confirmed
Error handling
The reviewed pages describe neither retries nor error paths for tasks and workflows.
Source 12Not yet confirmed
Sub-processors only partly readable
The Trust Center lists 31 sub-processors, of which 5 were visible when fetched. The data processing agreement and technical measures are marked "NDA" there.
Source 10
Can I use it?
With requirements
Europe & regional availability
Security page, terms, help articles and Trust Center reviewed. The Trust Center showed 5 of 31 sub-processors, and the data processing agreement and technical measures are marked "NDA". The terms give two different revision dates.
Reviewed scope: Hosted platform; model zone depending on selection
Reviewed on:
With requirements
Server location
The security page and help article name Frankfurt am Main, the security page also AWS. The pricing page names EU hosting in Berlin for the Section 203 licence, the article on innoGPT Standard its own infrastructure in the EU. The data flow article gives no more detail on the database location beyond Frankfurt.
Region, reason and source
Region covered: EU
Documented reason: The pages name different locations depending on the product part; no contract text was readable.
Official source · www.innogpt.de · Official source · docs.innogpt.de · Official source · www.innogpt.de · Official source · docs.innogpt.de
With requirements
Model providers and transfer
The help article names OpenAI through a direct API with standard contractual clauses, Anthropic through Google Vertex in Frankfurt, Google Cloud, Mistral, Perplexity as a US company and Replicate with processing in the USA. Admins choose zones such as EU/EU or Global/Global. According to the terms, choosing models outside the EU counts as consent.
Region, reason and source
Region covered: EU and USA
Documented reason: Location and transfer depend on the chosen model.
Official source · docs.innogpt.de · Official source · docs.innogpt.de · Official source · www.innogpt.de
With requirements
"Zero retention" with exceptions
The article promises no training but names retention for abuse detection: OpenAI up to 30 days, Mistral 30 days, Replicate one hour. It gives no period for Perplexity and Google Cloud.
Region, reason and source
Region covered: Per model provider
Documented reason: The headline is broader than the stated conditions; the provider lines are what counts.
Available
Data processing agreement
Workspace admins sign the data processing agreement under Art. 28 GDPR digitally in the settings; according to the article the contracting party is innoki GmbH. The text is marked "NDA" in the Trust Center and was not reviewed.
Region, reason and source
Region covered: EU/EEA
Documented reason: The way to sign is documented, the content is not.
Official source · docs.innogpt.de · Official source · trust.innogpt.de
Availability still unclear
Sub-processors and certificates
The Trust Center lists 31 sub-processors; visible when fetched were AWS, Clerk, Cohere, E2B and ElevenLabs, without purpose and location. It lists ISO 27001 and GDPR as "Compliant" and SOC 2 Type 2 as "In progress", with no certificate in the fetch. "GDPR compliant" is a provider claim.
Region, reason and source
Region covered: EU/EEA
Documented reason: 26 names and the documents not readable; no authority assessment found.
Official source · trust.innogpt.de · Official source · www.innogpt.de
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 workspace administration; self-operation only on the Enterprise plan
Sources reviewed:
Show all 3 points
Documented with conditions
Data access & team sharing
Apps must first be connected and authenticated. For Custom Actions the page names role-based approvals and an audit log. The reviewed pages do not describe the reach of individual connections in detail; according to the changelog SharePoint can be limited to selected sites.
Evidence · Evidence · Evidence
Memory and conversation continuity
1 of 6 criteria documented
Memories exist in the settings, per assistant and per project. According to the data flow article they feed in as context; the articles do not describe deleting and admin rules.
Memories for users, assistants and projects
Sources accessed:
Documentation by criterion
StoragePartly documented
Storage and scope
You set personal details and answer preferences in the settings, company context in the assistant's Memory tab, project details in the project. According to the article the platform for the data flow is in Frankfurt; it names no separate storage location for memories.
SelectionPartly documented
Selection and timing
You add entries or say "remember this" in chat. According to the documentation memories also update automatically, but it gives no rules for this. They can be imported from ChatGPT.
Edit & deletePartly documented
Inspect, edit and delete
Entries are in the settings and the assistant tab. The reviewed article does not describe editing and deleting.
Access boundariesDocumented
Agent and channel boundaries
There are separate levels for user, assistant and project. According to the changelog temporary chats do not access memories and do not save new ones.
RetrievalPartly documented
Retrieval
According to the article the platform adds chat history and, if relevant, content from memories to a request. It does not say how relevance is determined. Entries are split into smaller units to separate topics.
New conversationsPartly documented
Session reset and compaction
According to the article inactive chats are deleted after 180 days. The reviewed pages say nothing about compacting sessions or resetting memories.
Cost and effort
4 plans · Prices as of 5 November 2026
Prices and included usage
Pricing checked
Prices per user per month plus VAT. The trial lasts 7 days without payment details and includes only standard models.
| Plan | Price | Included |
|---|---|---|
| Business | EUR 25 per user per month; 1 month free with annual payment | All features, team collaboration, up to 1,000 users, SSO/SAML at extra charge. |
| Business Max | EUR 109 per user per month | More room for premium models for individual users. |
| Section 203 licence | EUR 25 per user per month; annual only for 1 to 20 licences | For professional secrecy holders: confidentiality commitment, according to the page EU hosting in Berlin and no model training, enabled by the team. |
| Enterprise | Custom, from 1,000 users | Self-hosting or on-premise, dedicated support manager, individual SLAs. |
- According to the page standard messages are unlimited, premium models use a budget.
- Third-party price information was not used.
Setup
Test for 7 days without payment details. For agents you create assistants, connect apps or MCP servers and have admins set up Custom Actions.
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: Hosted operation by the provider; self-hosting on the Enterprise plan according to the pricing page
Sources reviewed:
Editorial assessment
innoGPT verifiably offers tools, subagents and scheduled runs. The documentation, however, sits mainly in the changelog, and error handling remains open. Check the hosting zone and model provider for each model yourself, because location, retention and transfer depend on them.
Show all 6 points
Documented
Files and working state
According to the help article conversations are stored in the database in Frankfurt, and inactive chats are deleted after 180 days. Admins can configure deletion periods. Projects exist independently of individual users; removed members lose their chats permanently.
Evidence · Evidence · EvidenceDocumented with conditions
Starting further runs
The home page names schedule and event as triggers. The changelog names scheduled tasks, running agent tasks with the status "Running" and native triggers from Microsoft 365. The automation page with "Coming soon" does not match this.
Evidence · Evidence · EvidenceDocumented with conditions
Human intervention
According to the changelog routines always ask for approval before actions outside innoGPT. Local MCP tools can be tied to a confirmation. For Custom Actions the agent chooses by itself according to the page, and it names no approval per call.
Evidence · EvidenceOpen in reviewed sources
Resuming after failures
According to the changelog paused tasks and workflows stay stopped. The reviewed pages do not describe retries or resuming aborted runs.
Evidence
Evidence and open questions
20 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, security page, terms, help articles, changelog and Trust Center reviewed. The Trust Center showed only parts. Feature descriptions come from the provider; 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.
Provider
Inno KI GmbH, based in Vechta, offers the AI platform innoGPT for companies.
- Operator
- Inno KI GmbH
- Organization
- Provider
- Website
- www.innogpt.de
Changes to this profile 1
AI platform from Vechta with chat, over 50 models, assistants with tools, scheduled tasks and workflows in public beta.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.