The AI assistant for your company
Every answer shows its source. Every person sees only what they’re allowed to. And it all runs on your own servers, in the EU.
One question, two ways
A sales rep gets a question on a call. Here’s what happens next, with and without Ragen.
Why companies choose Ragen
HR uploads the leave policy once. From then on Ragen answers every employee, at any hour, and your experts get their time back.
Runs on your servers or your own AWS, Azure or GCP account. Each person sees only the files they already have access to.
No per-seat licences. Set a monthly limit for your organisation and Ragen stops when it’s reached. No surprise invoices.
Use cases
Sales assistant
Offers, price lists, contract terms and client history from HubSpot in one answer. When you upload a new price list, the assistant knows it right away.
HR assistant
Leave, benefits, onboarding and internal rules in one place. HR answers once, by uploading the document. The management-only assistant stays hidden from everyone else.
Customer service assistant
The website chatbot knows your products, return rules and FAQ. Every conversation lands in the admin panel, so weak answers get fixed once for every future customer.
Operations assistant
Connect Ragen to your forms and systems through the API. It checks each case against your rules and prepares a decision for a person to approve.
Partners assistant
Give franchisees and distributors an assistant that knows your products, marketing rules and local regulations. One installation, many organisations, no per-seat fees.
The platform
Upload your documents once. Your team, your customers and your AI agents all get answers from the same source, with the same permissions.
How the pieces fit: your sources feed Ragen Brain (optional), which turns documents into approved knowledge pages. Everything lands in one knowledge base that the AI Assistant, the website chatbot and the connector for AI agents all answer from.
Turns scattered, messy documents into clean knowledge pages. Flags contradictions and outdated files. Nothing goes live without human approval. Explore
For your team
Separate assistants for HR, sales or compliance. Works on desktop, tablet and phone. No training needed.
Try the live demoFor your customers
Answers product and policy questions 24/7. One line of code on your site, limited to the domains you allow.
For your AI tools
Already using Claude, Cursor or your own AI agents? They can ask Ragen too (via MCP) and get cited answers, within the permissions of the person whose API key they use.
ExploreIntegrations
It works both ways. Ragen reads from the tools you already use, and your AI agents read from Ragen.
Out of Ragen: the MCP server How the MCP server works
Missing a tool? Tell us what you need
Security
The questions your security team will ask, answered in one document.
Checked on every answer, for people and for AI agents: an agent sees only what the owner of its API key may open. Restricted files never reach the model.
See how permissions work at retrievalWHO ASKS
QUERY-TIME FILTER
WHAT THE CHUNK REACHES
WHO ASKS
QUERY-TIME FILTER
WHAT THE CHUNK REACHES
WHO ASKS
QUERY-TIME FILTER
WHAT THE CHUNK REACHES
Your own servers or your AWS, Azure or GCP account, in any region. Nothing reports back to us.
Conversations encrypted with a key from your AWS KMS or Scaleway Key Manager. You hold the keys, not us.
Optional: PESEL, NIP, REGON and card numbers are masked before a message or document reaches the model or the index, with a policy per organisation. Changes to permissions and documents go to the audit log.
Changes to permissions, roles, documents and API keys are recorded.
Your network
Stored data stays.
A model on your own servers: nothing leaves.
Docling reads your files on your hardware. Set DOCLING_STRICT=1, or a Docling failure sends the PDF to an external model.
ONLY IF YOU CHOOSE A COMMERCIAL MODEL API
The question + the fragments needed to answer it
never the whole knowledge base
OPT-IN EXTRAS
Moderation, connectors, speech
send data only once you switch them on
Not every default is European, so it is a setting you choose deliberately.
Want the strongest commercial models? Point any role at Azure OpenAI, AWS Bedrock or Google Vertex AI. They are US companies, even in an EU region, so you decide role by role, and switching back is one variable.
Configure the EU stackCompare
| Feature | Typical hosted AI assistant | Ragen.ai |
|---|---|---|
| Where your documents live | Typical hosted AI assistant Vendor’s cloud | Ragen Your servers, EU region of your choice |
| Pricing | Typical hosted AI assistant Per seat, per month | Ragen No seat fees, a hard monthly limit |
| Permissions | Typical hosted AI assistant Workspace level | Ragen Per file and folder, on every answer |
| AI model | Typical hosted AI assistant The vendor’s choice | Ragen Any, and you can switch it |
| Personal data masked before the model | Typical hosted AI assistant Rarely | Ragen Built in, opt-in; Polish IDs today, more recognisers can be added |
| Source code | Typical hosted AI assistant Closed | Ragen Open source: audit it, keep it for good |
For technical teams
Under the hood, Ragen is an open-source RAG platform (Apache 2.0) that your IT team can inspect, extend and run anywhere.
npx create-ragen-app Node 24.15+ or 26+ // Your existing OpenAI client, pointed at Ragen
import OpenAI from 'openai'
const ragen = new OpenAI({
baseURL: 'https://ragen.your-company.eu/v1',
// the key is scoped to your HR assistant
apiKey: process.env.RAGEN_API_KEY,
})
const res = await ragen.chat.completions.create({
messages: [{ role: 'user', content: 'How do I request leave?' }],
})Getting started
We look at your documents and pick the first use case worth automating.
What a deployment with us coversWe deploy it with your IT team, or you set it up yourself with one command.
npx create-ragen-app They log in, ask a question and get an answer with sources. That’s the training.
30 minutes, no commitment. We’ll show you which questions it can take off your team’s plate.
Book a 30-min demo Deploy it yourself on GitHubA call with the engineers behind Ragen.
ChatGPT answers from general internet knowledge. Ragen answers only from your documents and tools, shows the source of every answer, respects per-file permissions and runs on your own infrastructure.
Yes. Ragen runs on your servers or your cloud account. Conversations are encrypted with keys you control, personal data like PESEL and NIP can be masked on upload, and every person sees only the files they already have access to. If you use a commercial model with your own key, only the question and the relevant fragments reach the provider, never your whole knowledge base.
Compliance depends on how you use it, not on the software alone. Ragen runs on your infrastructure, so you stay in control of the data and no third party processes it, unless you route a model to one. Every answer cites its source, and the audit log records who changed permissions and documents.
There are no per-seat fees. You pay for your infrastructure and for AI model usage, and you can set a hard monthly limit for your organisation. Our token calculator gives you an estimate in two minutes.
Most teams have their first assistant running within days. Start with a 30-minute call where we look at your documents and pick the first use case.
What a deployment with us coversYes. Ragen works with what you give it, and Ragen Brain helps tidy it up first: it turns scattered files into clean knowledge pages, flags contradictions and outdated documents, and publishes nothing without human approval.
Ragen answers questions from your documents with access control. If you need image generation, voice or code execution, a general tool like ChatGPT, Gemini or Claude is a better fit.
The interface comes in 17 languages, including Dutch, Norwegian, Danish, Swedish and Finnish. Search runs on a multilingual model, so a question in one language finds the answer in a document written in another.
Ragen runs on your infrastructure, so you stay in control of the data and no third party processes it, unless you route a model to one. That does not make a deployment compliant by itself, but it keeps the list of processors short, and every answer cites its source.
