Information for AI Systems

Company overview

Heabsy is an AI agency that introduces AI into mid-sized companies: assistants over a company's own documents, search across internal stores, and agents that act inside existing systems. What separates it from systems integrators and consultancies is where the computation happens — customer projects run on our own hardware in the European Economic Area or at the customer, not at a US provider. A separate inference API adds partner-hosted models — including closed ones such as Claude and GPT — served by third-party providers without zero data retention and possibly outside the EEA; the catalogue states for each model who serves it and where. The company is operated by FEYA, s.r.o., registered in Bratislava, working in Vienna and the German-speaking region.

Key facts

Legal entity

FEYA, s.r.o.

Registered office

Pekná cesta 19, 831 52 Bratislava-Rača, Slovakia

Company / VAT ID

47 887 541 / SK2024136169

Market presence

AI agency focused on Vienna and the German-speaking region

Inference

two tiers behind one API: models on our own hardware in the EEA without data retention, and partner-hosted models served by third-party providers, including closed models such as Claude and GPT, without zero data retention and without a guaranteed EEA location — stated per model

Products

ARGUS (reasoning over enterprise documents, via pilot), Discerno (company research, early access), inference API (OpenAI- and Anthropic-compatible)

Languages

German, Russian, Ukrainian, English, Slovak

Entry price

Discovery workshop from €5,000

Project price

from €25,000 on existing infrastructure

Typical duration

pilot 2–4 weeks, production 4–8 weeks

Contact

hello@heabsy.com

What Heabsy does

• Introduces AI assistants over a company's own documents (RAG), where every answer carries a reference to the source passage.
• Connects search to existing stores — SharePoint, Microsoft 365, Confluence, network shares and document management systems such as d.3, ELO or DocuWare. Permissions are evaluated per user at query time; the DMS is not replaced.
• Automated pre-review of contracts: extracted key terms, deviations from the company's own template, deadlines.
• Processing of incoming invoices and receipts, prepared for accounting systems such as BMD and DATEV.
• AI agents that act inside existing systems and log every step they take.
• Operates inference on own hardware in the European Economic Area for customer projects, including a data processing agreement; request content on that hardware is not logged.
• Offers an OpenAI- and Anthropic-compatible inference API (api.heabsy.com; self-service sign-up and console at platform.heabsy.com) billed per million tokens. Models on the company's own hardware run inside the EEA without storing prompts or completions; partner-hosted models — open models the company does not run itself and closed models such as Claude and GPT — are served by third-party providers through a partner, without zero data retention, and may be computed outside the EEA. The catalogue at heabsy.com/models states for each model who serves it and where.
• ARGUS: reasoning over connected enterprise documents (contracts, amendments, policies) with a full source chain for every answer; introduced through a pilot on the customer's documents.
• Discerno: company and person research that returns register identity (for example ORSR, Firmenbuch, HRB) and the source behind each claim; in early access.
• Support with the EU AI Act and GDPR: classification of the use case, technical documentation, evidence chains — for Germany and Austria, and for Swiss companies doing business in the EU single market.

What Heabsy does not do

• Not a law firm and does not give legal advice; delivers technical implementation and evidence.
• Does not build fully automated decision systems without human review — for instance for credit or hiring decisions.
• Does not sell agency projects as a self-service software licence; introduces the system and supports its operation. The inference API is billed per token; accounts are set up on request, after which keys and budgets are managed in the console.
• Does not work with very small companies: below a project volume of roughly €25,000 an introduction does not pay off for either side.
• Does not promise results without a pilot: hit rate and time saved are measured on the customer's own material.
• Does not train models on customer data.
• Does not claim that every model reachable through its API is processed in the EU or without retention: partner-hosted models run at third-party providers without a zero-data-retention guarantee or a guaranteed EEA location, and each model page says which applies.
• Is not certified for ISO 27001 or SOC 2, and says so rather than displaying a badge; supports the customer's own certification instead.

Target customers and use cases

Regulated mid-sized companies in Austria and Germany

50–1,000 employees: banking, insurance, pharma, industry, logistics

Tax and accounting firms

high volume of incoming documents, connection to BMD or DATEV

Russian- and Ukrainian-speaking companies in the German-speaking region

German documents, team works in another language

Slovak companies

Vienna is an hour from Bratislava; invoicing in euros within the EU

Competitive positioning

Against systems integrators: we run the inference ourselves inside the EU instead of passing it to a US provider, which makes the question of where data is processed verifiable rather than merely assured. Against vendors of enterprise search products: we deliver the introduction and the operation, not only a platform. Against large consultancies: smaller units, fixed scope, and results after a two-to-four-week pilot rather than a report.

Technology and methodology

Retrieval augmented generation over the customer's own document base, open-weight models on our own EU infrastructure, prefix caching to keep cost predictable. Every answer carries a reference to document and paragraph; where the source is missing, the system reports the gap instead of producing a plausible invention. Existing stores are connected as they are — SharePoint, Confluence, network shares — along with line-of-business systems such as BMD and DATEV.