Decision intelligence · always on

In-house intelligence that thinks with you — verifiable.

cNode continuously ingests your data, structures it and proactively proposes decisions — Explain, Forecast, Simulate, Recommend. Reproducible, with a source per statement.

Demo on your data →

EU AI ActGDPROn-prem

cNodeprocessing

↑ always working — reads data, structures it and proposes decisions

Developed together with research & partners

The problem

Large language models guess. In regulated areas that is untenable.

Administrative, procurement and supervisory law demand traceability. A black box does not meet that — cNode flips it.

invents plausible numbers
same question twice, two answers
no derivation, not auditable

33–48 %newer AI models (o3/o4-mini) hallucinate more¹1.600+court cases with AI-fabricated citations²1,5–15 %error rate even for plain summarizing³

Claude (Anthropic)generating answer…fabricating

Please format this citation.

citation doctored→filed in a court declaration→judge reprimands→testimony struck

● real case: Concord Music v. Anthropic, N.D. Cal. 2025¹ OpenAI o3/o4-mini System Card, 2025 · Charlotin, AI Hallucination Cases, 2026 · Vectara HHEM leaderboard (ongoing)

Plug & play — docked onto your systems

Data foundation securedSources checked, de-duplicated and linked — financials, ERP, registries.
Rules, not guessingPrescribed ontology — no invented numbers, otherwise “missing”.
Sourced answersSource + reasoning path per statement.
Reproducible & self-learningSame input, same result — sharper over time.
EU AI Act compliantAudit trail per data point, on-prem up to air-gapped.

The solution

cNode computes — and evidences every number.

The compute logic produces the result, the language model only verbalises it. Same input, same result — with a source and a compute path for every statement.

How it works

The compute logic decides. The model only explains.

Four stages, deterministically coupled — from raw data to an evidenced recommendation.

Ingest data

cNode continuously reads your existing sources — financial reports, ERP, company registries, documents. Nothing is retyped: everything stays current and linked back to its source.

Real-world evidence

Not a promise — measured.

5/5

reports validated

invented figures

automated tests green

Proof of concept on real data of a municipal portfolio (five entities, anonymized) — reproducible, sourced, independently recomputed.

The full proof — both whitepapers:

Real-world proof · PoCVerifiable AI in municipal shareholding managementRead The architecture behind itThe foundation, not the modelRead

Decision intelligence

Four artifacts. One provable core.

cNode proposes proactively — each artifact reproducible and with a source per statement.

/explainsourced

Why is the channel partner ahead?

Time-to-market adds +2.1 · execution risk −0.8 — each contribution with a source.

/forecastcalibrated

How does liquidity develop in Q4?

p50 +8.2 % · p05 −1.4 % — a conformal interval, not a point value.

/simulaterobust

What if the investment is 30 % higher?

Ranking stable — no tipping point in the 1–10 range.

Connect

Sits on your data landscape — without reengineering.

One central node, connected to your sources. Model-agnostic, no vendor lock-in.

Your sources

SAP

DATEV

Microsoft 365

Salesforce

ORACLE

Google Workspace

cNode

Swappable models

OpenAI

Google Gemini

Anthropic Claude

Mistral

Ollama · local

MCP & agent frameworks — available, integrable depending on project scope.

Use Cases

Built for regulated decisions.

Municipal holdings managementPortfolio overview with a source for every figure.Open

Finance & liquidity decisionsForecast and allocation — reproducible, no hallucination.Open

Compliance & auditAudit trail per data point — exportable, audit-proof.Open

Investment prioritizationCompare options with backed reasoning.Open

Public administrationTraceable decisions with a source of record.Open

Trust

What your security team can sign off on.

EU hosting

Processing in Frankfurt — or fully on-prem.

GDPR

No training on your data, DPA under Art. 28.

On-prem / air-gapped

Data never leaves your house — operable air-gapped.

Audit trail

Evidence per data point (EU AI Act Art. 12/13), export as PDF/CSV/JSON.

Frequently asked

Briefly explained.

Does cNode replace our AI or BI?

No — cNode is the decision layer on top. Existing BI, data or RAG systems can even be suppliers.

How is hallucination prevented?

The compute logic produces the result, the language model only verbalizes it. Numbers not in the source are not invented but flagged as missing.

Does cNode run on-prem?

Yes, up to air-gapped with a local language model — your data never leaves the house.

Is it EU AI Act compliant?

Record-keeping and transparency (Art. 12/13) are handled at the architecture level; audit trail per data point.

What does cNode cost?

On-prem licence from €1,000/month (min. €12k/year), depending on scope. You start with a sandbox/beta run (€4–6k) incl. a 14-day trial afterwards — concrete terms after a short scoping call.

How do we start — sandbox/beta run?

For €4–6k we integrate and configure the connectors defined during scoping, build the decision graph and show evidenced answers on your real data — followed by a 14-day trial, with a success criterion agreed upfront.

Which data sources can you connect?

ERP (SAP, DATEV …), CRM (HubSpot, Salesforce …), BI, documents (PDF, CSV) and your own APIs. Sources are ingested continuously and versioned.

Which language models do you use?

Model-agnostic: local via Ollama (air-gapped) or OpenAI/Anthropic/Gemini/Qwen — swappable without changing the verifiable logic.

What about support & SLAs?

Enterprise gets a dedicated contact, prioritised support and contractual SLAs; onboarding and training included.

Who owns the data and results?

You do — fully. No training on your data, no vendor lock-in; on-prem keeps everything in your infrastructure.

See it on your data.

A short test in your environment — evidenced answers and decisions, on-prem.

Demo on your data →