This notice explains how c:node uses artificial intelligence, which models are involved, how we label AI results and where people decide. It supplements the Privacy Policy and the Terms of Use.
1. Who we are and our role
Creativate Technologies GmbH, Lahnstraße 65, 60326 Frankfurt am Main, Germany, email: privacy@creativate.tech
- Provider of the AI system c:node (Art. 3(3) AI Act): we develop c:node and place it on the market under our name.
- Deployers (Art. 3(4) AI Act) are the organisations that use c:node for their staff. For the situation report demo on c-node.ai and the trial environment, we are the deployer ourselves.
- General-purpose AI models: c:node uses language models from other providers (Google, Anthropic, OpenAI, Mistral) and open models we run ourselves. We do not train our own foundation models and do not modify these models in a way that would make us the provider of such a model.
2. What c:node does with AI
| Function | What happens | Role of AI |
|---|---|---|
| Chat and answers | answers questions with sources from your content, connected services and the c:node Graph | language model writes; statements are supported by sources and checked after writing |
| Drafts (emails, documents, appointments, tasks) | creates drafts that an authorised person edits and approves | language model writes the draft; nothing goes out without approval |
| Calculation engine | calculates with your tables and files (e.g. costings, variances) | calculation is deterministic without AI; AI only chooses the module, proposes column mappings (you confirm) and explains the result |
| Officer lookup | names the officers of a company registered in the commercial register, with source and historical status | no AI; rule-based from register data |
| Routines | runs recurring tasks on a schedule on your behalf | as chat; the result is always a labelled draft that must be approved |
| Templates for municipalities and public administration (e.g. council resolution drafts, funding outlines, reports) | creates structured document drafts | text parts by AI; figures exclusively from the calculation engine |
| Cockpit / steering | shows key figures from tasks, accounting, CRM and calendars | figures rule-based; optional summary labelled "AI draft – please check" |
| Situation report demo on c-node.ai | summarises public information about a company | language model (Google Gemini) summarises |
| Voice input | converts speech to text | local speech recognition model on our servers |
3. Models used
| Provider | Models (as of October 2026) | Use | Location |
|---|---|---|---|
| Google (Gemini API) | Gemini Flash, Gemini Pro | default in the trial environment and basic plans, classifying questions, web search | USA / worldwide |
| Anthropic | Claude (Haiku, Sonnet, Opus) | Professional/Enterprise, fallback | USA |
| OpenAI | GPT models | selectable in Professional/Enterprise | USA or EU |
| Mistral AI | Mistral Small/Large | selectable, "EU only" policy | EU |
| Own/open models | e.g. Qwen via Ollama, Whisper (speech), embedding models | on our servers or at the customer | EU or at the customer |
You can see which model wrote an answer in the application under "Sources & reasoning". Organisations can require that only models processing in the EU are used. We do not make your data available to any provider for training purposes (see Privacy Policy, sections 3 and 4.8).
4. Risk classification under the AI Act
In our assessment, c:node in its intended use is not a high-risk AI system under Art. 6 and Annex III AI Act. c:node is an assistance system for research, documentation, calculation and drafting; it does not make decisions about people. The transparency obligations under Art. 50 AI Act and the AI literacy obligation under Art. 4 apply to c:node.
| Function | Classification (proposal) | Reason |
|---|---|---|
| Chat, answers, drafts, routines | limited risk (Art. 50) | interaction with people, text generation; results are checked and approved by people |
| Calculation engine | minimal risk | deterministic calculation; AI only for mapping proposals and explanation |
| Officer lookup | not an AI system in the narrow sense | rule-based query of public register data |
| Templates for municipalities and public administration | limited risk (Art. 50) | drafts of documents that people check, change and adopt; no assessment of persons and no decision on benefits |
| Cockpit | minimal risk | rule-based figures |
| Situation report demo | limited risk (Art. 50) | AI summary of public information about companies |
Uses not intended. c:node must not be used to decide on persons in the areas listed in Annex III AI Act or to materially determine such decisions, in particular not for selecting or evaluating applicants or employees, assessing creditworthiness, deciding on public benefits or for biometric procedures. Prohibited practices under Art. 5 AI Act are excluded.
5. Labelling and transparency (Art. 50 AI Act)
- You are talking to an AI. c:node presents itself as an AI system at sign-in, in the application and in the demo; answers appear as AI answers.
- Evidence instead of claims. Supported statements carry their source; statements without evidence are labelled "General assessment". Calculation proposals carry "AI suggestion – please check", AI summaries in the cockpit "AI draft – please check", results of routines the note "Created by routine … · draft – please check".
- Exported documents (DOCX, where enabled for the organisation) carry the note "created with AI support (c:node)" while in draft; after approval, who approved the document and when is recorded.
- Shared answers carry the note "Created with c:node — AI that backs every statement with its source".
6. Human oversight
- Nothing external without approval. Every email, appointment, document and task in connected services is presented as a draft and only executed after explicit approval by an authorised person. Sending an email is a second approval step. There is deliberately no "approve all" function.
- No approval by machines. Neither a language model nor an interface (API, MCP) nor a routine can approve; this is only possible in a signed-in c:node session.
- Review note. Supported answers carry a human-oversight note: it is a supported decision proposal, and a person makes the decision.
- No automated decisions within the meaning of Art. 22 GDPR.
7. Logging and traceability
- Every approval is stored with person, time and changes made.
- For every answer an audit record with sources, evidence status and checksum can be exported (JSON, CSV, HTML).
- Per model call we log provider, model, task and consumption without content (400 days).
- Routine runs, block list changes and MCP calls are logged without content.
- Before every release, automated evaluations check, among other things, evidence quality and approval safety.
8. AI literacy (Art. 4 AI Act)
Our staff who develop or operate c:node or advise customers are trained on how the AI used works, its limits and risks. Organisations using c:node are, as deployers, responsible for the AI literacy of their own staff; we provide help pages, this notice and training on request.
9. Limitations
AI results may be incomplete or wrong, even when sources are given. Sources from public registers may be outdated (e.g. historical register status). c:node does not replace legal, tax or financial advice. Check results before using them.
10. Contact and reporting problems
Please report questions, indications of incorrect or biased results and incidents to privacy@creativate.tech or via the feedback icon in the application. We review every report.
Last updated: 21 October 2026 (version 2.1.0) Creativate Technologies GmbH