The little helper that stands by you with every click

The helper knows the current module and current screen: current booking mask, opened customer record, open risk inventory or document being edited. Questions like 'What does cost center mean here?' or 'Which value belongs in this field?' are answered directly, without you having to explain the context. Answers are short and tailored to the screen, no long Wikipedia article, no theoretical excursions. For complex masks the helper recognises the active tab or active accordion and adjusts the answer. References to further wiki pages or training content are delivered automatically, you click directly through to the deeper material if desired. For repeated questions the helper remembers your experience level and shortens answers accordingly for power users.

The helper doesn't just answer, it acts. Example in the finance module: 'Suggest me a journal entry for this document', the helper fills account, cost center and amount directly in the form. In CRM: 'Create a follow-up appointment in two weeks', the helper creates the appointment and links the customer record with all relevant notes. In the HR module: 'Prepare the contract renewal', the helper opens the matching template and fills in personnel master data. In the ISMS module: 'Link this measure with the matching Annex A controls', the helper suggests the mappings. You only click 'accept' or correct. Actions are reversible, every change is documented in the audit trail with the note 'suggested by AI helper, confirmed by user' including time stamp and user ID.

When you correct a helper suggestion, the system learns from the correction. Example: you book documents from a specific supplier always to a special cost center, the helper takes that over after three to five corrections automatically. Learning is tenant-isolated: corrections from your tenant never influence other tenants, technically via Postgres Row-Level Security and logically via few-shot example selection. No training on your data in the model itself, everything stays in your tenant. Learning happens via few-shot examples that are loaded per query from your tenant, without your data influencing the model weights. Power users can see learning entries and maintain manually, e.g. remove unwanted patterns or mark especially valuable corrections as 'sticky'. On employee change, personal learning entries are anonymised or transferred to the successor.

Hover over a field or click the helper icon, the helper explains what belongs in. For regulatory terms (e.g. 'GoBD-compliant storage', 'TOMs under Article 32', 'record of processing activities under Article 30 GDPR') it delivers the standard-compliant definition plus the reference to the specific mask with examples from your tenant. Explanations are in informal address, German, without legalese or marketing language. Power users can hide explanations, beginners see them by default, default per user and per module maintainable. For mandatory fields the helper explains why the field is mandatory and what consequences leaving it empty would have (e.g. 'GoBD violation if document type not filled'). Examples for valid values are shown automatically, based on past entries in your tenant, anonymised and without cross-tenant data.

Recurring tasks the helper takes over completely on request: monthly travel-cost document import from Outlook mailbox, weekly status report to the team with drill-down links, reminder for contract renewals 30 days before expiry with prepared renewal template. You define the routine once in natural language ('Send me every Monday at 8 a.m. an overview of overdue measures in the ISMS'), the helper suggests the configuration, you confirm. Routines run via the workflow engine, every run is documented and traceable in the audit trail including trigger and result. For errors (e.g. missing data source or changed permission) you get a notification with concrete error cause and solution suggestion. Routines are versioned, you roll back to previous configurations if needed.

The helper uses Mistral Small 2 for fast answers and Mistral Large 2 for more complex suggestions, both EU-hosted at Mistral in France. Multi-tenant isolated on Postgres with Row-Level Security, the helper only sees data of your tenant and only those the logged-in user has access to. Tenant context is validated on every query, no accidental cross-tenant leak. No OpenAI, no data outflow to the USA, no training on your data, contractually assured via DPA addenda with Mistral. Model selection automatically by task complexity, you only notice that answers take a bit longer or shorter depending on the question. For KRITIS tenants a 'Mistral-only EU proxy' mode is also available, all queries run anonymised through an EU proxy in Frankfurt, no US cloud share in the stack.