FreyaLabs

Same intelligence. Different place.

AI that knows your company —and never leaves your premises.

Freya answers questions from your own documents with citations and prepares drafts. On your hardware or in a closed EU environment — your people always approve.

Free intro call (30 min)How it works

No obligation · Reply within 2 business days

Your premises · local or EUDiagram: manuals and PDFs, office files, Microsoft 365, Google Workspace, Confluence and Jira, and other systems come together in FreyaKI inside your premises. Below, Freya answers questions from different departments with sources.FreyaKIManuals & PDFsOffice filesOther systemsMicrosoft 365Google WorkspaceConfluence & JiraDiagram: manuals and PDFs, office files, Microsoft 365, Google Workspace, Confluence and Jira, and other systems come together in FreyaKI inside your premises. Below, Freya answers questions from different departments with sources.FreyaKIManuals& PDFsOffice filesOthersystemsMicrosoft365GoogleWorkspaceConfluence& Jira

Protection class 0 · Public: Approved services possible

What torque applies to the flange on pump P-40?

FreyaKI

45 Nm, cross pattern in two passes. Replace the gasket every time it is opened.

Source: Manufacturer manual P-40, p. 112

Fictional example
  • GDPR-compliant operation
  • On-prem or in the EU
  • Humans approve
  • Company based in Germany

Why AI often fails to land in operations

Three hurdles that frequently stall AI projects in mid-sized companies.

Chatbots do not know your business

Without your manuals, quotes, and files, answers stay generic — or are simply made up.

Sensitive data must stay inside

Customer data, prices, and HR files do not belong in third-party clouds. So AI stays unused.

Answers without evidence cost time

If it is unclear where a statement comes from, someone ends up checking everything by hand.

EU AI Act · Data Act

The deadlines are already running

The EU AI Act’s first obligations apply since February 2025, the Data Act since September 2025. The strictest tiers follow through 2028 — preparation starts now.

Milestones

High fine risk

Next tier in384days

Today: 13/11/2026

  1. AI Act: prohibited AI practices, AI literacy duty for staff

    High fine risk. Fine: up to €35M / 7%

  2. Data Act applies

    Fine: up to €5M

  3. AI Act enforceable; Bundesnetzagentur is AI supervisor

    High fine risk. Fine: up to €15M / 3%

  4. High-risk AI, including HR and recruiting

    High fine risk. Fine: up to €15M / 3%

  5. High-risk AI in products and machinery

    Fine: up to €15M / 3%

When things go wrong

€35.3M

H&M · Hamburg · 2020

Employee health and private data was collected and visible to dozens of managers.

Freya principleProtection classes and access rights

Hamburg DPA (3 Oct 2020)

Refund

Deloitte Australia · 2025

An AI report contained fabricated sources and a fake court citation.

Freya principleEvery answer cites its source

ABC News (8 May 2025)

Risk — how Freya is built

AI Act: up to €35M or 7% of turnover (prohibited practices), up to €15M or 3% (other obligations); for SMEs the lower amount applies. Data Act in Germany (DADG): up to €5M.

In the intro call we clarify where you standStart the check

Not legal advice. As of November 2026.

Sources and notes

How Freya brings your knowledge together

Scattered documents and systems become one knowledge base — with answers that name their source.

Prepared, not just stored

Freya reads PDFs, scans, and office files, splits them into searchable passages, and stores them in collections with a fixed protection class.

Diagram: PDFs and scans, office files, and business systems are read by Freya, split into passages, and stored in the knowledge base by protection class, such as Internal, Confidential, or Strictly confidential.Knowledge baseRead & splitPDFs & scansOffice filesBusinesssystemsInternalConfidentialStrictly confidentialDiagram: PDFs and scans, office files, and business systems are read by Freya, split into passages, and stored in the knowledge base by protection class, such as Internal, Confidential, or Strictly confidential.Knowledge baseRead & splitPDFs & scansOffice filesBusinesssystemsInternalConfidentialStrictly confidential

Found, filtered, cited

Freya combines keyword and semantic search, respects your protection classes, and backs the answer with sources.

Diagram: a question runs through keyword and semantic search, filtered by protection class. The result is an answer with sources.QuestionKeyword searchSemantic searchClass filterAnswer withsourcesDiagram: a question runs through keyword and semantic search, filtered by protection class. The result is an answer with sources.QuestionKeyword searchSemantic searchClass filterAnswer withsources

What Freya does for you

From knowledge base to approval — under your control.

Answers from your knowledge

Freya searches your documents and answers with a citation. You see right away where every statement comes from.

Operations you control

On your own hardware or in a closed EU environment. Per protection class you decide which data may go where.

Drafts, not solo runs

AI prepares quotes, replies, and review reports. Your staff approve — nothing goes out unchecked.

Use cases

Four typical workflows — and what Freya prepares for your team.

Trades

Quote from a customer request

  1. IncomingA customer emails a bathroom renovation request with three photos.
  2. Freyachecks site measurements, price list and free install slots.
  3. ResultDraft quote your master craftsman reviews and sends.
Freya · Draftawaiting approval
“Site measure missing. Install from week 6 — material confirmation still pending.”
Fictional example

Mechanical engineering

Fault at an existing customer

  1. IncomingCustomer reports: “Our P-40 pump is losing pressure.”
  2. Freyauses the maintenance manual, history and parts stock.
  3. ResultReviewable service proposal with spare part and sources.
Freya · Draftawaiting approval
“Check variant → align readings → approve procedure. Diagnosis open.”

Finance

Completing a client file

  1. IncomingNew documents arrive for a financing request.
  2. Freyacompares forms and attachments and flags contradictions.
  3. ResultGap list with sources for your case handler.
Freya · Draftawaiting approval
“Proof missing; field on form p. 2 conflicts with attachment A.”
Fictional example

Public sector

Checking an application for completeness

  1. IncomingAn application with several attachments is submitted.
  2. Freyachecks it against the applicable rules and checklists.
  3. ResultDraft request for missing items — the decision stays with you.
Freya · Draftawaiting approval
“Please submit the missing proof and clarify the discrepancy.”
Fictional example

Use case · Mechanical engineering

From customer request to approved reply

A service request lands in the inbox. Freya agents prepare everything according to your process — your team only reviews and approves.

Fictional example
Diagram: a request from the email inbox passes through three Freya agents. The customer agent uses the CRM, the knowledge agent the knowledge base. Your team reviews and approves the draft; then the reply is sent and the CRM is updated.Freya agentsEmail inboxIntake agentCustomeragentKnowledgeagentCRMKnowledgebaseYour teamapprovesReply sentCRM updatedDiagram: a request from the email inbox passes through three Freya agents. The customer agent uses the CRM, the knowledge agent the knowledge base. Your team reviews and approves the draft; then the reply is sent and the CRM is updated.Freya agentsEmail inboxIntake agentCustomer agentKnowledge agentCRMKnowledgebaseYour teamapprovesReply sentCRM updated
We connect inbox, CRM and knowledge base via interfaces — tailored to your systems.

How the request is handled

  1. 1

    Request recognised

    An existing customer writes: “Our P-40 pump is losing pressure.” The intake agent identifies customer, machine and urgency.

  2. 2

    Customer in context

    The customer agent pulls contract, installed machine and maintenance history from the CRM.

  3. 3

    Solution with sources

    The knowledge agent finds the prescribed service process and the right spare part in the maintenance manual — and turns them into a draft reply with sources.

  4. 4

    Reviewed and approved

    Your service technician reviews the draft, adjusts it if needed and approves. The reply goes out and the case is documented in the CRM.

  • No searching across five systems
  • Every reply follows your process
  • Nothing goes out unreviewed
Discuss this workflow for your business

No obligation · Reply within 2 business days

Agent orchestration

One orchestrator, six specialists

Freya hands every task to the right agents, keeps track of the whole and brings the results together. You make the final call.

Fictional example
Diagram: Freya sits in the centre and steers six agents in two areas, core processes and foundation; the foundation also includes the connection to your systems. Different agents work together depending on the scenario, and you approve at the end. From enquiry to quoteCore processesFoundationWebsite enquiryMarketingSalesBack officeSupportKnowledgeProcessYour systemsYou approveDiagram: Freya sits in the centre and steers six agents in two areas, core processes and foundation; the foundation also includes the connection to your systems. Different agents work together depending on the scenario, and you approve at the end. From enquiry to quoteCore processesFoundationWebsite enquiryMarketingSalesBack officeSupportKnowledgeProcessYour systemsYou approve

From enquiry to quote

A prospect gets in touch through your website. Freya plans the work, hands it out and gives your sales team a finished draft.

That is what I want

How it runs

  1. 1Marketing captures the enquiry and qualifies the lead.
  2. 2Knowledge supplies prices, service descriptions and matching references — with sources.
  3. 3Sales drafts the quote and prepares the entry for your CRM.
  4. 4Process checks discount limits and approval rules from your process documentation.
  5. 5You review, adjust and approve the quote.

Agents not involved are dimmed. Your systems such as CRM, ERP and inventory are connected by arrangement.

Public AI chatbots or Freya?

The difference is not the model — it is the place and the control.

CriterionPublic AI chatbotsFreyaKI
Company knowledgeOnly what you paste inYour documents, searchable
Place of processingProvider servers, often outside the EUYour hardware or a closed EU environment
CitationsRarely traceableEvery answer with a source
Data pathsSet by the providerSet by you, per protection class
Approval before sendingNot built inA human approves every draft

Privacy built in, not bolted on

Useful AI knows your business — so the place of processing must fit.

  • Data path per protection class before every request.
  • No silent cloud paths — egress only via reviewed allowlists.
  • Exportable knowledge and swappable models — you keep control.

Five protection classes

  1. 0PublicApproved services possible
  2. 1InternalReviewed paths only
  3. 2ConfidentialNarrow data path
  4. 3Strictly confidentialVery narrow data path
  5. 4Especially criticalNo general AI access

A suggestion for your corporate policy. The higher the class, the narrower the data path.

From intro call to pilot

Start small. Draw a boundary. Test for four weeks.

  1. 01

    Intro call

    30 min · free

    You describe a task that costs time every week. We tell you openly whether Freya is a fit.

  2. 02

    Draw a boundary

    Defined together

    Together we define which data must never go to the cloud: customers, prices, staff, internal figures.

  3. 03

    Test for four weeks

    4 weeks · real cases

    With real cases, on-prem or in a closed environment. Then you decide on scaling up.

Book intro call

No obligation · Reply within 2 business days

Free

Get informed first, then decide

Our practical guide and the readiness check show you where you stand and what makes sense next.

Cover of the guide “KI im Unternehmen”

PDF · German edition, October 2026

AI in your company: the practical guide

16 pages, five steps from the first idea to a safe, verifiable pilot.

  • What AI does and where it runs
  • Protecting data, placing the law
  • Finding the right first task
  • 12 questions on AI readiness

You will receive an e-mail with a confirmation link (double opt-in). See our privacy policy.

12 questions · about 3 minutes

Readiness check

Twelve questions in six areas. You see right away whether you are at orientation, foundations, pilot-ready or scale-up.

  1. 0–6Orientation
  2. 7–12Foundations
  3. 13–18Pilot-ready
  4. 19–24Scale-up
Start the check

Frequently asked questions

The key answers up front.

What is Freya?

Freya is the AI workspace by FreyaLabs for small and mid-sized companies. Freya answers questions from your own documents with citations, prepares drafts and hands tasks to specialised agents. Freya runs on your hardware or in a closed EU environment, and your staff always approve the result.

Where does data run — cloud, local, or hybrid?

You decide per protection class and task. Sensitive processing stays on your side; approved content may go to vetted services — only via defined paths.

How does a pilot run?

You pick a concrete task, draw a data boundary, and test four weeks with real cases — then decide on continuation and scope.

What does it cost?

The intro call is free. We discuss scope and operating model there — then you receive a quote.

See all questions

Free intro call

Let's talk about your use case

30 minutes, no obligation. You describe a task — we show how Freya can solve it.

0 / 500 characters

We only use your details for your request.