Frequently asked questions

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.

For small and mid-sized companies in German-speaking markets whose knowledge sits in manuals, PDFs, spreadsheets and file shares — for example in industrial manufacturing, the trades, field service and administration.

An agent takes on one clearly defined task, such as capturing an enquiry, looking up knowledge or preparing a draft. Freya steers the agents centrally, brings the results together and presents them to you for approval. Systems such as CRM or ERP are connected by arrangement.

FreyaKI builds on RAGFlow as an open-source foundation (Apache 2.0), hardened for tenants, protection classes, and controlled operations. You keep export and model choice.

You choose the connection: local on your hardware (e.g. Ollama, vLLM) or approved cloud endpoints. We align with your performance and privacy reality.

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

Often a CPU with smaller models is enough to start. Larger models and high load benefit from dedicated hardware.

Suggested corporate policy (not a legal classification): 0 Public · 1 Internal · 2 Confidential · 3 Strictly confidential · 4 Especially critical (no general AI access). The highest contained class wins; also check personal data and statutory duties.

Different review dimensions: GDPR — lawfulness of personal data. EU AI Act — role and risk of the concrete AI use (transition dates include August 2026 application, December 2027 high-risk rules, August 2028 product-related high-risk — per our research notes). Data Act — access and portability for connected product data. Not legal advice — involve counsel when in doubt.

The AI Act classifies systems by risk and role (provider, deployer, importer). Prohibited practices and AI literacy apply since February 2025; enforcement and further tiers roll out through 2028. Which duties hit you depends on the use case — HR, safety, or products. Freya is built so approvals, protection classes, and citations reflect your operations; we do not replace legal assessment.

In short: 2 Feb 2025 prohibited practices and AI literacy; 12 Sep 2025 Data Act; 2 Aug 2026 stronger AI Act enforcement in Germany; 2 Dec 2027 many high-risk duties including HR; 2 Aug 2028 high-risk in regulated products. Details and exceptions are in the regulations — verify with counsel. The home page shows a timeline with a current “today” marker.

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

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