How the AI works
HelloBoss uses artificial intelligence (AI) in several places. This page states in plain words where that happens, how you can tell, and what the AI explicitly does not do. It fulfils the information duty under Article 50 of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689), which applies from 2 August 2026.
Where AI is used
- Speech recognition: voice messages from WhatsApp or Telegram are transcribed automatically.
- Translation: field messages are translated into the office language and office replies into the worker's language. Outgoing voice messages are read out by a synthetic voice — there is no human speaker.
- Classification: the AI suggests which site and which kind of message (sick note, missing material, completion …) a message belongs to.
- Summaries: short overviews for the boss and the office are generated from many individual messages.
- Reading uploaded files: line items, quantities and prices are extracted from quotes, bills of quantities and plans (PDF, spreadsheet, photo).
- Invoice and appointment drafts: a draft of what could be invoiced is derived from the reported progress.
How to recognise AI output
Wherever a person sees AI output, a notice sits next to it — a label on the content itself or a notice box above the whole section. In printouts, PDF reports and export files the same labelling appears as text with the marker [KI]. Every labelled area of the interface additionally carries a machine-readable marker (data-ki="ki-erzeugt"). A full C2PA provenance watermark is only required for existing systems from 2 December 2026 and is not implemented yet.
Humans decide
No AI suggestion takes effect on its own. Quotes are never sent automatically, invoices are never issued automatically, classifications can be changed at any time. Every AI output is a working aid and does not replace professional, commercial or legal review. HelloBoss contains no decision taken solely by a machine that would have legal effects for a person (no case of Article 22 GDPR).
What the AI does not do
- No rating, scoring or ranking of individual employees.
- No performance or behaviour monitoring, no covert surveillance.
- No emotion recognition and no biometric identification of persons.
- No decisions about hiring, dismissal, pay or promotion.
- No working-time tracking unless the customer company switches it on explicitly.
- No fabricated content about real persons, places or events.
Providers involved
- OpenAI, L.L.C. (USA) — speech recognition
- Anthropic PBC (USA) — text analysis and classification (Claude Haiku 4.5)
- OpenRouter, Inc. (USA) — routing, fallback model Mistral Small
- Microsoft Azure — synthetic voice (text to speech)
The application itself and the database run on servers in the EU. Details on data flows and legal bases: privacy notice.
Legal classification
HelloBoss is not a high-risk AI system within the meaning of Annex III of the AI Act. The decisive point is purpose limitation: the analysis serves site documentation, not the assessment of employees. If HelloBoss were used to evaluate, select or monitor people, that classification would change and the stricter high-risk obligations would apply.
Reporting mistakes
Wrong translation, misheard voice message, nonsensical suggestion: please write to hilfe@helloboss.dev. Data questions: datenschutz@helloboss.dev. Security vulnerabilities: report a vulnerability.
Version 2026-07-30. German version: /ki-transparenz (prevails in case of conflict).