EinCoreRAG
Private AI over your own documents — in the EU, with the model you choose.
Your staff and customers get answers from your own documents, with the source under every answer — hosted in the EU, with the model you choose. Personal data is masked before the model sees it, and every action is recorded in an audit trail that cannot be edited: entries cannot be changed or deleted, and are kept for a fixed period. Built for organisations where nothing goes live without a security review.
Early access: request access or a demo on this page. The free plan (1 document, 30 answers a month, no card) is part of it; for the larger plans we send an offer for your organisation.

EinCoreRAG
Features
Grounded RAG over your knowledge
Ingest PDF, DOCX, CSV, TXT or Markdown, or crawl your website; search returns cited answers from your content only. When the retrieved evidence is not enough, it is designed to abstain and hand the conversation to a human rather than guess.
Evidence, not assertions
A security review asks for artifacts, and the platform produces them: a GDPR Art. 30 record of processing your admin generates on their own, and — prepared with us — an EU AI Act transparency pack, dated, with a fingerprint per file and one over the whole pack, so any later change shows, a data-subject export that moves nothing until the requester proves their identity, and a sub-processor and residency register.
Bring your own model (BYOM)
Your own keys for Mistral AI, Anthropic, OpenAI or Azure OpenAI, AWS Bedrock in the EU, Groq, OpenRouter, or a model you run yourself on any server that speaks the common OpenAI interface, from Professional. You choose the provider and model and switch them without engineers; a same-provider fallback model covers transient errors.
Hosted in the EU, the model where you choose
Your documents, search index and audit trail are hosted in the EU. Where the model runs is your call: an EU provider with your own key, the EU regions of your cloud, or your own server — and then questions and retrieved passages stay in the EU too. Each customer's data is kept apart by two layers — the application on every query and the database itself — tested on 1,000+ workspaces, and Enterprise adds a separate search index per customer.
Personal data masked before a model sees it
Personal data is masked before it reaches the model — in the question and in the retrieved knowledge — and the answer is screened again on the way out. Not a filter bolted on at the edge.
Human handoff
Low-confidence answers move into your team's shared inbox, with SLA and business hours. If nobody picks a conversation up in time, the AI takes it back, tells the visitor so and lets your team know — nobody is left without an answer.
Every channel your customers use
A web widget that drops into your site, its styles kept separate from your page's, and answers in the customer's language — eight languages — labelled as AI (EU AI Act Art. 50) and with its sources. From Professional: WhatsApp, Messenger and Instagram, Slack, email, SMS, Zendesk and Intercom.
Built for your support team
A shared inbox where your agent takes over a conversation, with an agent copilot, canned replies and conversation summaries — and a weekly report of the questions it could not answer, so you know which document to write next.
Help centre, sign-in and spend under your control
A public help centre built from the articles you approve. Single sign-on and automatic user provisioning from Professional, a spend limit you set yourself, and data-subject export and erasure that run only after confirmation.
Cost and answer quality, measured
Cost per customer, a sample of answers checked against their sources, and a record of how each answer was produced. Audit entries cannot be changed or deleted.
- Data residency
- EUData residency
- Audit trail
- Tamper-evidentAudit trail
- Workspaces in the isolation test
- 1,000+Workspaces in the isolation test
- Every answer
- CitedEvery answer
For IT operations
Your runbooks and knowledge base, answering the service desk.
In the product
An answer with its sources, in your customer's language
What your customer sees on a phone: a straight answer, the sources behind it, and a person one tap away.

Labelled as AI
Every conversation opens with a notice that the customer is talking to an AI system.
Sources under every answer
The documents the answer drew on are listed beneath it, so a customer — or your team — can check it.
A person, one tap away
“Talk to a human” hands the conversation to your team.
How we start
From first call to deployment
A deployment on your own documents, prepared with you.
- 01
Scoping call
We map your use case, data sources, and compliance constraints — AI Act risk tier, DORA, GDPR.
- 02
Your own workspace in the EU
Under NDA: we load your documents into a workspace of your own — in the EU cloud, or on your own server, inside your network — and wire one real use case end to end.
- 03
Deployment and evidence
We deploy it on the agreed use case, against goals set at the start, with the audit trail and the documentation your security review needs.
Security & Compliance
Evidence your auditor can check, not promises
Every claim here maps to a mechanism we can show your auditor — nothing more.
Controls mapped to SOC 2 and ISO 27001
Our controls are mapped to SOC 2 and ISO 27001 evidence, so your auditor can trace requirements to working mechanisms. An external audit against SOC 2 and ISO 27001 is the next step on our roadmap.
Designed for the EU AI Act and GDPR
Designed for the EU AI Act and GDPR: an AI label and sources on every answer (Art. 50), and an AI Act deployer evidence pack (Art. 26 for high-risk use) per workspace; log retention set to your obligation.
Each customer's data kept apart
Separation enforced on every query by the application and by the database itself, tested on 1,000+ workspaces, with cached data kept apart per customer too — and in Enterprise a separate search index per customer.
Security firewalls
Prompt-injection detection and containment on every input and output, personal data masked before the model sees it, every write action held for human approval, and denial-of-wallet spend caps.
Tamper-evident audit trail
Cannot be edited, kept per customer, and enforced by the database itself. Evidence is structured for SOC 2, ISO 27001, and EU AI Act reviews.
EU-sovereign deployment
Hosted in the EU; model execution stays in the EU when you choose an EU provider, the EU regions of your cloud or your own server.
Where it runs
In the EU, on the model you choose
The EU cloud is opening in early access, from the free plan up — or Enterprise, on your own server.
- Early access
EinCoreRAG Cloud
Hosted in the EU, from the free plan up, with the model you choose.
- Your keys
Bring your own model
From Professional, connect your own keys — Mistral AI, Anthropic, OpenAI or Azure OpenAI, AWS Bedrock in the EU, Groq, OpenRouter, or a model you run yourself — and switch without engineers. With a provider outside the EU, questions and retrieved passages go to that provider.
- On your server
EinCoreRAG Enterprise
EinCoreRAG Enterprise is installed on your own server, inside your network — with our team. The model answers through the provider you choose, or through a model you run yourself inside your network.
Watch
Where it runs, and what stays in the EU
Twenty-five seconds on the deployment question — in the EU, grounded in your documents, every answer cited.
Listen
Hear the platform explained
A deep dive into the architecture, security, and sovereign deployment behind EinCoreRAG.
Inside EinCoreRAG
Platform Deep Dive
Who it's for
Built for organisations whose data must stay in the EU
Regulated organisations in the EU that need sovereign AI with the evidence to prove it.
- 🏦
Regional & cooperative banks
Your data hosted in the EU, the model where you choose, with controls mapped to DORA requirements.
- 🛡️
Insurers
An AI Act deployer evidence pack (Art. 26 for high-risk use) per workspace; log retention set to your obligation.
- 🏥
Hospitals & health groups
EU cloud, your choice of model, and personal data masked before the model sees it; Article 28 DPA and DPIA support included.
- 🏛️
Public sector
An EU-incorporated vendor, tender-ready, with local-language delivery.
Plans
Start small. Grow when the answers pay off.
Every plan runs the same platform: answers with their sources, personal data masked before the model sees it, EU hosting. Plans differ in how many answers they give and how many documents they know.
Starter
For a first assistant on your website.
- Website widget
- 900 AI answers a month
- 100 documents
- 2 team members
- Widget in 8 languages
Professional
For teams putting it in front of customers and staff.
- 3,000 AI answers a month
- 1,000 documents
- 5 team members
- All channels and integrations
- Single sign-on and automatic user provisioning
- Bring your own model (BYOM)
Business
For larger knowledge bases — arranged with our sales team, with onboarding.
- 10,900 AI answers a month
- 5,000 documents
- 10 team members
- Onboarding with our team
Scale & Enterprise
For large knowledge bases and regulated deployments — dedicated capacity, arranged with our sales team and priced per engagement.
Talk to us
The free plan: 1 document, 30 answers a month, no card. We may change or end the free plan; you are told in advance and can export your documents. During early access every plan starts with a short request on this page, and we send an offer for your organisation; Business and Scale are arranged with our sales team.
Enterprise
Bring us the AI project your compliance team said no to
EinCoreRAG Enterprise is installed on your own server, inside your network — with our team. We scope it with you on a call, around one real use case, with the evidence a review asks for.
Request a callLive demo
Watch it answer your questions, on a call we run.
Twenty-five minutes, screen shared, with the engineers who build it. No recording and nothing to install.
You set the subject
Which documents your people keep re-reading, and who asks you for answers from them — the regulator, a customer's security review, your own first-line support.
We ask it your questions, live
Every answer arrives with the source document it came from, on screen. You judge the answers, not a slide about answers.
You see it say "I don't know"
We ask something its knowledge base does not cover. It says so and offers a human instead of improvising. That behaviour is the reason this belongs in a regulated firm — a system that always has an answer is the one that eventually invents a wrong one in front of your auditor.
We end on your installation
Your documents, your choice of model provider, your deployment. That is a per-installation setting, not something you inherit from us.
Request access or book the demo
Tell us which documents you need answered and who asks you for them. We come back within one business day — with your access, or with two times for a call run by the people who build the platform.
What you send is used for one thing: arranging the call. Your name, email, company and question are stored in our own systems and emailed to our team. You are not added to any mailing list, and one email removes it.
FAQ
Frequently asked questions
Will it make up answers?
It is designed to answer only from your own indexed knowledge base and to cite the source. When the answer isn't in the sources, it says so and hands off to a human — so every answer can be checked.
Is each customer's data isolated?
Yes. The application and the database itself keep each customer's data apart on every query, tested on 1,000+ workspaces; Enterprise adds a separate search index per customer.
Our data has to stay in the EU — what are our options?
Your documents, index and audit trail are hosted in the EU. To keep model execution in the EU as well, choose an EU model provider with your own key or the EU regions of your cloud provider — from Professional up. For full separation, our team installs EinCoreRAG Enterprise on your own server, inside your network, with a model you run yourself.
What evidence do we get for auditors?
A per-customer audit trail whose entries cannot be changed or deleted, chained with SHA-256 so any change shows, and kept for a fixed period; an AI Act deployer evidence pack (Art. 26 for high-risk use) per workspace, with log retention set to your obligation; and data-subject export and erasure with a confirmation step. Structured for SOC 2 and ISO 27001 reviews.
Is there a free plan?
Yes, as part of early access: 1 document, 30 answers a month, no card. We may change or end the free plan; you are told in advance and can export your documents. For Starter, Professional, Business and Scale we send an offer for your organisation; during early access every plan starts with the request form on this page.
Can we use our own model?
Yes, from Professional: connect your own keys for Mistral AI, Anthropic, OpenAI or Azure OpenAI, AWS Bedrock in the EU, Groq, OpenRouter, or a model you run yourself on any server that speaks the common OpenAI interface. You choose the provider and model and switch them without engineers; a same-provider fallback model covers transient errors, and a spend limit keeps the bill where you set it.
Further reading
- Citations that can be wrong: grounding a RAG answer so the claim and the source actually matchA citation next to a sentence is a claim about that sentence, and four different things can be wrong with it. The failure taxonomy, why a 0.95 faithfulness score can sit on a wrong answer, why LLM-as-judge evaluation cannot be trusted to grade itself, and what to build instead — including the refusal path most systems never ship.
- EU-sovereign AI stack: when Mistral + Qdrant beats a US LLMA decision framework for choosing between an EU-resident AI stack and a US frontier LLM, weighing data residency, GDPR, and the AI Act.
- Air-gapped AI: what zero egress actually costs, and what it does not buyAir-gapped is an operating commitment, not a network setting. The eleven things a normal inference stack phones home about, how to prove zero egress instead of declaring it, the update clock you inherit the day you unplug, and the obligations an air gap does not remove.
Ready to see it working for you?
Tell us what you need and we come back within one business day with the next step.
Built by an EU-incorporated senior team — 20+ years in enterprise delivery.