Sternenstaub.AI · AI that stays in-house
Your enterprise AI: GDPR-compliant, affordable, in your hands.
Sternenstaub.AI makes your company's knowledge usable via chat, securely in-house, with a source for every answer and without runaway licence costs. Instead of handing your data to big cloud corporations, the AI runs where your data already is.
Sound familiar?
AI benefits, yes, but not at any price.
Data sovereignty vs. cloud lock-in
Company and customer data must not leave the house uncontrolled. GDPR, trade secrets and trust are at stake, yet the big AI suites presuppose the cloud.
Costs out of control
A licence per user per month quickly gets expensive and is hard to plan, especially for mid-sized companies.
Your context is missing
Standard chatbots don't know your systems, processes and terminology and have no access to your internal knowledge.
Sternenstaub.AI closes exactly this gap: an AI that runs in your environment, knows your sources and backs up every answer.
The solution
Your knowledge, usable via chat, with evidence.
Assistant with source citation
Answers from your own documents, transparently backed up instead of made up. Technically a RAG assistant (retrieval-augmented generation).
Connect knowledge sources
BookStack wiki, SMB file shares, EspoCRM, Microsoft SharePoint, local file uploads (incl. image and audio recognition) and targeted web research, freely extensible.
Traceability
Knowledge graph and answer tracing show how systems and documents relate and what the AI answers from, with each source's influence directly adjustable.
Security & data sovereignty
Self-hosted (on-prem or EU cloud), encryption, strict tenant separation, permission-aware answers and optionally access via VPN only.
In detail
Features that make the difference.
The assistant that names its sources
Employees ask in natural language and get answers from their own company knowledge. Below each answer, the sources used appear with relevance. So it's always clear what a statement is based on.
- Answers with evidence, multiple chat sessions
- Select or deselect sources, thematic bundles (“spots”) and work environments
- Attach files directly (private, visible only to you); knows date and time, optionally current web info
Knowledge graph: connections at a glance
Sternenstaub.AI links your systems, folders, documents and topics into an interactive knowledge graph with semantic connections. The folder structure of your file shares and SharePoint libraries is fully represented, arbitrarily deep and with a document count per folder. In addition, folders can be tagged: if a folder carries e.g. “customer, customer data”, the AI searches there first for customer questions, including subfolders.
- Systems · folders · documents · topics
- Tag folders: search follows your filing structure
- Drag, zoom, full screen
Answer tracing: what does the AI answer from?
Ask a question and see which documents, systems and folders are drawn on, and with what relevance, down to the individual passage. Relevance is based on semantic similarity (vector embeddings). If an answer isn't right, adjust each source's influence directly (from “off” to “strong”) or force individual documents. The ranking is recalculated instantly and affects real answers.
- Relevance per source · cross-system
- Adjust each source's influence directly
- Force individual documents for an answer
Training & quality, without data science
Your experts adapt the assistant themselves, without AI specialists. Define persona and tone, maintain a glossary, store binding answers and learn from 👍/👎 feedback. A quality dashboard shows answer quality over time.
- Persona · glossary · binding Q&A
- Feedback loop
- Quality measurable
Answers that arrive on their own
Recurring questions run automatically, for example “Summarise the new wiki entries every Monday.” On a schedule (hourly to yearly) or by event trigger, such as when a document is added. A calendar view across day, week, month and year keeps track; the result lands in the history, as a chat session or via webhook into your own systems.
- Time- or event-driven
- Calendar: day · week · month · year
- Delivery to the history, as a chat session or via webhook
Connect your systems (extensions)
Knowledge sources are connected as extensions at the click of a button, without rebuilding. Alongside wiki, file server, CRM and SharePoint, also local file uploads with automatic image and audio recognition. Large holdings are synced in batches in the background, even tens of thousands of files, with visible progress, interruptible and resumable at any time.
- SMB · BookStack · EspoCRM · SharePoint · Web · local uploads (image/audio)
- Even tens of thousands of files in the background, interruptible
- Credentials encrypted
Access rights, spots and API
You control who sees which sources: under the principle “open until secured”, a source is visible to everyone until an access group is placed on it. After that, only its members see it, down to folder level. “Spots” bundle sources thematically (e.g. “onboarding” or “sales”) with their own behaviour. And via tokens you embed the assistant through a chat API into your own applications or intranet.
- Rights per group, down to folder level
- Spots: thematic source bundles
- Tokens: chat API for your own applications
Security, data sovereignty & operation
Sternenstaub.AI runs either as a GDPR-compliant cloud solution or on-premises directly on your own server. That way your data stays where you want it. Answers are permission-aware: each person only receives evidence from sources they are otherwise allowed to see. Sign-in runs via your central identity provider (Keycloak/OIDC), optionally with two-factor authentication, and on request all knowledge-source access runs mandatorily via VPN (fail-closed: no tunnel, no connection).
- GDPR-compliant cloud or on-premises at your site
- Permission-aware answers, sign-in via Keycloak/OIDC, optional 2FA
- VPN enforcement (optional), choice of language model (local or online)
A clear difference
Why Sternenstaub.AI, not the big cloud AI suite.
| Criterion | Sternenstaub.AI | Large cloud AI suites |
|---|---|---|
| Data sovereignty & GDPR | Data stays in-house or in the EU, no third-party use | Data in the provider cloud, control limited |
| Cost model | Plannable; local models usable free of charge | Licence per user/month, quickly expensive |
| System integration | Your systems: wiki, file server, CRM, SharePoint, web, local uploads | Primarily the vendor's own ecosystem |
| Traceability | Source citation per answer + knowledge graph, influence per source adjustable | Often a black box without provenance |
| Rights & visibility | Permission-aware answers down to folder level | Depends on the provider's ecosystem |
| Adaptability | Persona, glossary, binding answers, custom connectors | Standardised, limited |
| Operation | Self-hosted: on-premises or EU cloud, multi-tenant | Operated in the provider cloud |
Comparison is categorical (e.g. versus Microsoft Copilot or ChatGPT Enterprise); the feature set varies by provider.
Your path to Sternenstaub.AI
Productive in a few steps.
Getting to know & PoC
Try it out in a few days with your own systems.
Connection
Connect knowledge sources, adjust behaviour and glossary.
Rollout
Roll out multi-tenant and scale cost-consciously.
Operate flexibly
On-premises or German cloud: you decide.
Operate either on-premises (e.g. Apple Mac Studio or Linux GPU server) or in a German GPU cloud (e.g. Hetzner, IONOS), meaning GDPR-compliant with data centres in Germany. For a typical mid-sized customer (25 employees), plannable from a few thousand euros.
Frequently asked
Good to know.
Does our data leave the house?
No. Sternenstaub.AI runs on-premises or in an EU cloud; your data stays under your control and is not used to train third-party models.
Which AI models are used?
Either a free local language model (LLM) or a powerful online model, depending on need and budget, set by eyeworkers.
Do we have to “train” an AI model?
No. Curated knowledge (glossary, binding answers) and the feedback loop improve the answers immediately, without data science.
How much effort is the start?
A proof of concept is possible in a few days with your own systems.
Does every employee see all content?
No. Answers are permission-aware: each person only receives evidence from sources they are otherwise allowed to see. Sources can be secured per access group, down to folder level.
Can we embed the assistant into our own applications?
Yes. Via tokens, Sternenstaub.AI provides a chat API for your own portals, the intranet or other applications.
Is it GDPR-compliant?
Yes. Data sovereignty, encryption, strict tenant separation and optional VPN enforcement are core to the design.
Get to know us, no strings
Let's talk about your use case.
In a free demo we show Sternenstaub.AI with your own data and clarify which knowledge sources deliver value fastest.
Prefer to write? kontakt@eyeworkers.de
Thank you!
Your enquiry is on its way.
We usually get back to you within one working day.