Skip to content

Product

What demi is, who it is for and what it actually does.

10
What is demi?

demi is the AI platform for industrial companies. demi sits alongside your existing systems, brings your data together in one Data Layer, turns your org chart into permissions and only then lets AI work: traceable, and with each person's permissions. Nothing is replaced.

The platform
What problems does demi solve?

AI knows too little to be useful, or may do too much to be safe. demi solves both in three layers. The Data Layer is designed to bring your data together into a digital twin of the whole company. The Permission Layer turns your org chart into rights. In the Intelligence Layer, people and their AI agents do routine work together, such as answers to technical inquiries and tender analyses.

The platform
Who is demi built for?

For industrial companies with 20 to 5,000 employees, first for makers of technical B2B products, for example in fluid technology, sensors and mechanical engineering. We start in Germany, Austria and Switzerland.

All industries
Why not just one more AI app?

Because every AI app with its own data is one more silo. And a chatbot without your data guesses: it does not know your data, cites no source and checks no numbers. demi brings the whole company together: every app on demi works on the same Data Layer, with the same permissions. No new silo.

Which apps run on demi?

demi Spec reads tender packages and returns a list you can check, every value with its source. demi Ask answers technical enquiries from your own product knowledge. With demi Chat you ask your company's knowledge, within your permissions. demi Agents do recurring work on a schedule, in your role. demi Projects turns a goal into a plan and hands out the tasks. And with demi Talk you dictate instead of typing.

Intelligence Layer
What work does demi actually take on?

Routine work along the whole order: structuring enquiries from the inbox, answering technical customer questions, preparing quotes from similar orders, matching tenders against the product range, preparing orders from emails as checked order drafts and pre-filling 8D reports from case data. The use case library shows typical tasks by department and industry.

All use cases
Where should I start?

With a task that comes up often and takes a lot of manual work today, such as sorting inquiries or answering questions about delivery dates. Cases like these are a good start, because every company has them.

All use cases
Do I need all three layers?

Yes, they build on each other. The Data Layer is designed for your data landscape, the Permission Layer mirrors your org chart as rights, and only then does the Intelligence Layer work on top. You still start with one case, and every further case uses the same foundation.

How is demi different from a CRM?

A CRM stores what your team did. demi uses that data and the data from all your other systems: people and their AI agents do routine work with it, such as answering technical inquiries. Your CRM stays. demi works with each person's rights and writes back only after sign-off.

Which languages does demi support?

demi works in German, English and many other languages, for your team and your partners.

Data and systems

How demi works with your system landscape without replacing anything.

10
Do our systems stay, and does demi write to them?

Your systems stay. demi writes back only after sign-off. The Data Layer connects ERP, CRM, documents and email. Nothing is replaced, and you can go back at any time.

The Data Layer in detail
Which systems does demi connect to?

demi connects to any system, from SAP and Oracle to the specialist software of your industry: Microsoft 365, Google and Dropbox, ERP (e.g. SAP), CRM (e.g. Salesforce, HubSpot), PLM and CAD, MES, DMS and file servers. Your system is not on the list? Show it to us in the demo and we build the adapter with you.

See the demo
What is the Data Layer?

The first of demi's three layers. It is designed for ERP, CRM, PLM, DMS, mailboxes and drives, to bring them together into one data set: one record per customer, part and order, and every value keeps its source. Nothing is replaced.

The Data Layer in detail
What is the digital twin of the organisation?

A linked model of the whole organisation in demi's Data Layer: people, customers, products, orders, machines and documents as nodes, their relationships as edges. Every value keeps its source, and demi stores its own results there too. So the twin grows with every task.

Go to the glossary
How current is the data in demi?

demi works with the current state of the data that is connected. How quickly changes such as a new order, a new contact or the new version of a data sheet show up, we clarify in conversation.

What happens when two systems contradict each other?

What sits in several systems, demi merges into one record, and every value keeps its source so you can check it. If, say, a quote and the master data disagree, the agent shows the contradiction instead of silently taking one value.

Do we have to migrate data?

No. No migration, no new system to maintain: the Data Layer is built for ERP, CRM, PLM, DMS, mailbox and drives and works with what already runs. Your systems stay as they are.

Who owns the data in the Data Layer?

Your data belongs to you: export at any time as JSON or through the API, and for at least 30 days after the contract ends. What applies to model training on your data is governed by the DPA (§ 3 para. 3).

Can we build our own software on demi?

Yes. The Data Layer is open and has a standardised data model on which you use AI and build your own software. demi is headless: your team builds its own applications and workflows on it, through the API, MCP and the command line.

Go to the documentation
Will demi replace our systems one day?

First next to the old systems, then in their place: in the long run demi wants to hold the organisation's data itself, so you can retire old systems at your own pace. Today your systems stay, and demi writes back to them only after approval.

Permissions and approvals

Who sees what, who approves and what gets logged.

10
Can demi's AI see or change what a person may not?

No. An AI agent never has more rights than the person it works for. If an agent has read content that may carry instructions from third parties, every send, publish or share to the outside needs a person's click. demi checks every right on the server, not in the AI model. Each company sits in its own tenant, deleting is blocked for agents, and a log names the person, the agent and the time.

The Permission Layer in detail
Where do the rights come from?

From your org chart: departments, teams and roles. demi never widens rights on its own, and private documents stay private. That holds for people and their agents.

The Permission Layer in detail
Where does demi check permissions?

On the server, not in the AI model. So the AI cannot get around the check. Every query checks every record on its own: whoever may not open a file today will not see it in demi either. That holds for people and their agents.

When does an agent need a person's click?

If an agent has read content that may carry instructions from third parties, every send, publish or share to the outside needs a person's click. Deleting is blocked for agents, and actions by external AI clients need their own sign-off.

May an AI agent delete anything?

No. Deleting is blocked for agents. The log names person, agent and time and is protected against change and deletion.

Who may see the log?

The log names person, agent and time and is protected against change and deletion. It is there to trace what happened, not to monitor performance. Your company decides who may see it, for example in a works agreement.

How is the permission model built?

Permissions follow from relationships: tenant, organisation, organisational unit, team, project and document tree. They are inherited along this graph, and targeted exclusions such as private documents override an inherited right. On top comes a role concept with graded levels.

All technical details
How is different customers' data kept apart?

Every customer lives in its own tenant. The database itself separates the companies' data (row-level security).

With which rights does an agent work, technically?

Agents act with short-lived, delegated tokens bound to task, person and tenant. Destructive operations are blocked for agents, and actions of external AI clients need their own approval.

All technical details
What do we tell the works council?

If an agent has read content that may carry instructions from third parties, every send, publish or share to the outside needs a person's click. The log serves traceability, not performance monitoring. Who may see it is up to your company, for example in a works agreement.

AI and models

Which models work, where they run and why answers come with sources.

10
Which AI models does demi use?

demi is not another AI model. It is the layer AI models work on. The AI gateway demi HALO picks a fitting model for each task. Agents and image generation work with models in EU data centres; other regions only with your company's approval. The models are interchangeable, with no change on your side. The value sits underneath: in your data, your rights and the work your team does with them.

All AI models
Does demi work with any AI model?

demi is tied to no single model: one gateway, several providers, swappable. Connected today are, among others, Claude models, GPT models and open-weight models such as Qwen, Llama and gpt-oss, plus models for images, search and speech recognition. New models come in behind the same interface, with no change on your side.

All AI modelsModels in the docs
What is demi HALO?

demi HALO is demi's AI gateway. It picks a suitable model for each task. Agents and image generation work with models in EU data centres; other regions only with your company's approval. If there is no fallback route in your zone, the request stops. Models are swappable, with no change on your side.

All AI models
Do the AI models run in the EU?

Agents and image generation work with models in EU data centres, for example in Frankfurt, Paris and Milan; other regions only with your company's approval. For special tasks such as video, global models are available too.

All AI models
Do you train AI models on our data?

What applies to model training on your data is governed by the DPA (§ 3 para. 3). demi runs in EU data centres. Agents and image generation work with models in EU data centres; other regions only with your company's approval.

Can the AI make up figures?

Figures are calculated, not guessed: from your data, by fixed rules instead of estimates, and answers name their sources. If, for instance, quote and master data disagree, the agent shows the conflict instead of silently taking one value.

How do I know an answer is right?

The answer cites its sources. Every number points to where it comes from, such as deliveries in the ERP, reports in the DMS or contracts in the CRM, and to the rule it was calculated by. You check the answer instead of hunting for it.

How do people and AI agents work together in demi?

A person and their AI agent do the same work with that person's rights, and the log names both. A prompt, a message or an event starts a workflow or an agent. The result is a task, a request, an order or an answer with its sources. demi runs in the web app, as a desktop app, by voice and in external AI clients.

Can demi also handle tasks that take hours?

Yes. demi runs long-running tasks from start to finish, such as a tender analysis across many documents. A person orders it in the demi app, AI agents do the routine work on their behalf, every step is logged, and the result lands in the workspace with its sources.

Does my voice stay on my computer?

With demi Talk, yes: you speak, demi Talk writes, and speech recognition runs on your own computer (macOS, Windows, Linux). Audio you use in the demi app is processed by a speech recognition service from our sub-processor list.

Security and compliance

GDPR, hosting, encryption, certificates and what the contract says.

11
Is demi secure and GDPR-compliant?

demi runs in EU data centres. Agents and image generation work with models in EU data centres; other regions only with your company's approval. The data processing agreement under Art. 28 GDPR is part of every contract, our TOMs are public, and what applies to model training on your data is governed by the DPA (§ 3 para. 3). Where a provider has a location in the USA, EU standard contractual clauses or an adequacy decision apply. The sub-processor list names every provider and its location.

Where is our data stored?

In the EU. demi runs in EU data centres with Germany as the primary site; backups are kept separately in a second data centre. The sub-processor list names every provider with its location.

Sub-processor listGo to security
Which certificates do the data centres hold?

The provider of the primary site is certified to ISO/IEC 27001, 27017, 27018 and 27701. These are the provider's certificates for its services, not demi's. Which providers we use, for what and where, is in the sub-processor list.

Sub-processor listGo to security
Is demi certified to ISO 27001 or SOC 2?

Since November 2026 our internal processes and standards have been going through a pre-certification against Annex A of ISO/IEC 27001:2022. Our goal is certification to ISO/IEC 27001, followed by a SOC 2 report. Information security is a binding development and operations standard for us: a threat model and security checklist per specification, automated tests and security checks on every change. Every finding is versioned as a numbered specification and can therefore be audited.

TOMs under Art. 32 GDPRGo to security
How is our data encrypted?

At rest with AES-256, in transit with TLS 1.2 or newer. Backups are encrypted too.

How do our people sign in?

Through a central login with OpenID Connect, with single sign-on via your identity provider. Everyone can turn on multi-factor authentication, with a TOTP app or a passkey. The application database stores no passwords, and API keys are kept only as hashes.

What happens in a data breach?

We report in two stages: a first report within 48 hours at the latest, details within 72 hours of becoming aware (DPA § 6).

Which service providers process our data?

The complete list of all sub-processors is public and contractually binding, every provider with its location. We announce material changes in advance, with a right to object and a special right to terminate. If a provider has a location in the US, EU standard contractual clauses or an adequacy decision apply.

Sub-processor list
How do I report a security vulnerability?

By email to security@godemi.com. Our vulnerability disclosure policy describes what we need and how we handle your report.

Report a vulnerability
Can we get an extract of the log?

Yes, on a justified request, for your tenant. Every authenticated API request is logged with person or system, tenant and time, changing requests with their result.

We have special security or compliance requirements. Can you cover them?

Talk to us. We go through your requirements with you point by point, based on our data processing agreement, our TOMs and the sub-processor list.

Pricing and getting started

What your IT needs, how the first day works and what demi costs.

9
What does demi cost, and how is the price made up?

That depends on your plan: we agree the base fee, seats and credits with you. Credits measure the AI work demi does: your people's questions and the cases demi handles. Answer five short questions about industry, tasks and systems, and you usually get a personal offer within 24 hours on working days.

Request an offerOpen the ROI calculator
How quickly is demi up and running?

The first case runs in about a day, read-only at first. Nothing is replaced. You add more systems step by step, and demi writes back only after your sign-off. If you want help at the start, our services cover it.

How the first day worksServices / FDE
What does our IT need?

Not much: for each system demi should read, one technical user with read rights, write rights only where demi should write back after sign-off, and single sign-on via your identity provider (OpenID Connect). For each task the AI sees only what the person may see. Your data belongs to you: export at any time as JSON or through the API.

All technical details
Do we need to start an IT project for this?

No. Nothing is replaced. The first case runs in about a day, read-only at first. We do the setup together with you.

How does the first day work?

In three steps. Connect: demi works as a technical user with read-only rights. How data from your ERP gets to demi, we clarify in conversation. Set permissions: whoever may not open a file today will not see it in demi either. Work: the first case runs on real data, for example a spare-parts enquiry, answered from your systems, with its source.

How the first day works
What does demi bring in hours and euros?

Do the maths yourself: in the ROI calculator you enter your numbers, the assumptions sit right next to them, and so does the calculation. Free and without obligation.

Open the ROI calculator
Do you support us through the rollout?

Yes, as a partner, if you want. Our Forward Deployed Engineers work with your team, on site or remote, over months: map the processes and pick the first cases, clarify the data sources, set rights from the org chart, build cases and agents with your people, train, measure the gain and hand over. Our own team does the work, without subcontractors: the same people who build the demi platform.

Services / FDE
How do I see demi on my own case?

In a demo: in 20 minutes we show you demi on a case from your industry. Bring data together, set permissions, let AI work.

See the demo
What if my case is missing?

The library shows typical tasks, not all of them. Because every case runs on the same three layers, a new one usually needs no project of its own. Tell us about it in the demo.

API, MCP and CLI

demi in Claude, ChatGPT, Copilot, in the terminal and in your own software.

8
Where can our team use demi?

Wherever it works. demi is headless: interfaces dock onto the demi Console, the demi web app, MCP servers for external AI clients, a command line and API, a desktop app and a voice interface.

Go to the documentation
Is there an API?

Yes. The demi API at api.godemi.com is described as an OpenAPI 3.1 document. You sign in with OAuth or with a personal API key that you create in the app under Settings: with the rights read, write or run agent, and with an expiry date. A key never acts with more permissions than its person.

API docs
Is there a command line?

Yes, the demi CLI for macOS, Linux and Windows, installed with one command: curl -fsSL https://download.godemi.com/cli/install.sh | sh. You sign in with demi auth login, in CI with an API key. You edit pages as Markdown: demi docs pull, change, demi docs push.

CLI docs
What is MCP, and what does demi do with it?

The Model Context Protocol is an open protocol that AI agents use to call external tools. Through demi's MCP server, external AI clients such as Claude, ChatGPT or Copilot work with your demi, and every action of an external AI client needs its own approval.

MCP docs
Which AI clients does demi work with?

With any MCP client that supports Streamable HTTP and OAuth. Add https://mcp.godemi.com/mcp as a custom connector and sign in with your demi account; you choose your workspace while doing so. No download needed.

Connect any MCP client
How do I connect demi to Claude?

In Claude under Customize › Connectors › Add custom connector: enter a name and the URL https://mcp.godemi.com/mcp, then sign in with your demi account. In Team and Enterprise organisations an owner adds the connector once, then everyone clicks Connect. In Claude Code: claude mcp add --transport http demi https://mcp.godemi.com/mcp.

Connect ClaudeConnect Claude Code
How secure is access through MCP?

You sign in with OAuth 2.1 and PKCE, with your demi account, and the client works with your permissions. Every tool is marked read-only, changing or destructive, changing actions need their own approval, and deleting, ordering lead reports and creating API keys are blocked from AI clients. External AI clients stay off until your IT enables them for your tenant.

Security of the integration
Are there webhooks or real-time events?

Changes arrive as an event stream (server-sent events) through the API, and on the command line with demi events subscribe. There are no webhooks today.

API docs

Company

Who is behind demi, where we come from and how we hire.

11
Who is behind demi?

demi Technologies GmbH from Munich, founded by Philipp Munsch (sales, marketing and customer relations), Tobias Wolff (product, engineering and technology, CTO) and Niklas Löckel (operations, finance, legal and customer integrations). All three earned their master's degrees at the Technical University of Munich and come from industry and industrial consulting: Simon-Kucher, BMW, Porsche Consulting and MHP.

About us
Is demi a TUM Spin-off?

Yes. demi is a TUM Spin-off of the Technical University of Munich, grown out of the TUM Venture Labs and part of the UnternehmerTUM ecosystem. What we research is on the research page.

Go to research
What research does demi build on?

demi comes from the Technical University of Munich: the founders studied there, the CTO in Information Systems, and demi grew out of the TUM Venture Labs. Our research directions: the data layer and merging records from many systems, permission models for AI agents, traceable agents and how to evaluate them, distributed synchronisation and an information-theoretic view of data quality.

Go to research
Who supports demi?

The Federal Ministry for Economic Affairs and Energy (BMWE) funds demi in the EXIST-Forschungstransfer programme (EXIST Transfer of Research), co-financed by the European Union through the European Social Fund Plus (ESF Plus). The European Space Agency (ESA) admitted demi to its Business Incubation Centre (ESA BIC Bavaria). demi grew out of the TUM Venture Labs, completed the UnternehmerTUM XPLORE Market Pioneer programme, is a member of the NVIDIA Inception Program and takes part in the Elevator Ignite accelerator of the European Digital Innovation Hub AI5innovation.

About usAll supporters
Do demi's supporters hold equity in the company?

No. None of the public bodies, universities, programmes or companies listed on our “Supported, funded and financed by” page holds equity in demi Technologies GmbH. Public funding comes as grants; the programmes add coaching, networks and cloud credits.

All supporters
Where is demi based?

In Munich, at the MTZ near the Olympic Park. demi is developed in Germany and hosted in the EU.

Who already uses demi?

Industrial companies from component, machine and plant engineering have signed a demi subscription. demi Spec, for example, is in use at a pump maker with packages of dozens of PDFs and at a machine builder with single-machine enquiries.

About us
Is demi hiring?

Yes. We are building the team for Europe's AI operating system for industry, in Munich. Roles in engineering, sales and customer success are open right now, all on the careers page. If no role fits, send us a speculative application.

Go to careers
How does applying to demi work?

A personal-fit interview with one of the founders comes first, then a skills interview. On top, there is an AI challenge for everyone: you automate a real task with current AI tools, on the command line and on Linux, and show how independently you work. In sales you run a test sales call and solve a strategy case; in engineering there are coding challenges without and with AI and a review of AI-written code. Finally, you meet us in Munich.

The hiring process
Does demi itself work with AI?

Yes. We do our own work with AI agents: coding, sales, operations, hiring. One person and their AI agent, two actors in one role.

Go to careers
Is demi the same as demi.ai?

No. demi is the platform of demi Technologies GmbH from Munich, at godemi.com: the European AI Operating System for industrials. demi.ai is a different product from another company, an AI assistant for the Mac. The two are not related.

Your question is not here?

Search pages, use cases, industries and documentation, or talk to us.

See demi on your own case

In 20 minutes we show you demi on a case from your industry: bring the data together, set permissions, let AI work.

Your configurator

No use cases yet

Your configurator is still empty

Add a use case from any card or case page and see what it brings per year, with the calculation. Free, no sign-up, no commitment: try it out and change any number.

Browse use cases