The SLMers appliance (pronounced slammers)
Kiwacht-Edge running on a €500 mini PC.
What’s inside the SLMers box?
• Mini PC (AMD Ryzen 5, 16GB RAM, 512GB SSD)
• Pre-loaded with Kiwacht-Edge Docker image
• Printed quick-start guide (no cloud setup)
A Small Language Model (SLM) is a lightweight AI, typically under 4 billion parameters, far smaller than LLMs (large language models) like OpenAI’s GPT (175B+).
SLMs use a fraction of the energy, they can run on a standard office PC or even mobile phones, while LLMs require massive data centers.
On its own, an SLM can already summarise, translate, or assist with code, but with platforms like Kiwacht-Edge, an SLM powers real-world business workflows - voice alerts in retail, invoice validation for accountants, DVSA form automation, or license-compliant asset handling in creative agencies.
The future of AI isn’t bigger models, it’s smaller, faster, sovereign, independent intelligence which runs offline, respects GDPR, and stays under your control.
Europe cannot compete in the AI race between the US and China by building bigger models. Europe must choose a third path, being sovereign, sustainable, and guided by humans (EU AI Act).
It is not a question of whether artificial intelligence will transform the world, this is already happening. The real question is, who will shape the transformation? Billionaire CEOs? Nation states? China or the US?
The main winners of an AI-driven economy will not be workers or consumers, but the owners of models, data, and computational infrastructure. Europe must not become a data colony in this new world order.
The owners of superintelligence like OpenAI, Gemini could declare that your data is kept private, yet affect your life negatively by aggregating the financial decisions you make on the stock market.
Kiwacht-Edge keeps intelligence in your office.
Microsoft is document-centric, Apple is app-centric and Kiwacht-Edge is workflow-centric.
We encourage local programmers or business owners to build custom SLM solutions by writing rules.json by hand using KiWacht-Edge. JSON is our workflow engine.
Typical agentic models use JSON as a machine-to-machine protocol. We use JSON because it’s transparent, editable, and universal. Anyone, either the developer or the business owner can open rules.json and see exactly what the system will do.
Neuro-symbolic AI is a hybrid AI approach and is considered a "third wave" of AI.
This isn’t theoretical AI. It’s sovereign, auditable, human-guided automation.
Europe must reduce dependency on US and Chinese cloud AI. SLMs offer a sovereign path - no data export, no vendor lock-in.
more72% of EU SMEs cite data privacy as a top barrier to AI adoption (Eurostat, 2024). SLMs enable GDPR-compliant AI by design - no cross-border data flows.
moreSovereign AI doesn't require nuclear-powered data centres. It runs on a €500 laptop.
moreA 100% offline, on-prem AI engine powered by SLMs. Your data stays yours. Never used to train third-party models
moreA live demo of email triage using AI - GDPR-compliant, zero storage. Try it at kiwacht.de.
more
Google isn’t just adding a convenience feature.
It’s normalising the idea that your most sensitive business communications, client briefs, contracts, financial disputes, creative IP, should be processed by a US-based AI on foreign servers, as a default.
Even if Google claims it “doesn’t train on your data,” the act of sending every email to its cloud infrastructure creates three irreversible shifts:
1. OpenAI collects virtually everything you input. Everything becomes OpenAI’s training data by default (unless you opt out).
2. OpenAI shares data with third parties globally. Even under GDPR, they rely on Standard Contractual Clauses (SCCs), which have been repeatedly invalidated by EU courts (e.g., Schrems II).
Schrems II (Case C-311/18) is a landmark 2020 Court of Justice of the European Union (CJEU) ruling that invalidated the EU-US Privacy Shield framework, deeming it insufficient to protect EU citizens' data from US surveillance.
Large Language Models rely on infrastructure controlled by US and Chinese tech giants, creating strategic vulnerability for European organisations.
Small Language Models (SLMs) are compact, efficient, and locally deployable and enable true digital sovereignty. They run on European hardware, under European control, compliant with GDPR from day one.
This isn't about rejecting innovation. It's about reclaiming agency in the AI era.
× CloseMost cloud-based AI systems route sensitive communications through servers outside the EU and therefore can create legal risk under GDPR and eroding trust.
Kiwacht-Edge runs entirely within your infrastructure. No data leaves your control. No third-party access. No hidden training.
This isn't compliance as an afterthought - it's sovereignty by architecture. For the 72% of European SMEs held back by privacy concerns, SLMs aren't just an option - they're the only viable path forward.
× CloseEurope is investing billions in Small Modular Reactors (SMRs) to power energy-hungry AI infrastructure. But this path deepens dependence on massive, centralised systems.
Europe’s projected SMR investment (2023–2033) - to power AI that may not need such scale.
Kiwacht-Edge is an alternative: high-value intelligence without high-energy cost.
By using efficient SLMs, we deliver sovereign AI that is green, affordable, and accessible - without building new reactors.
× CloseTrust in AI erodes when users don't know where their data goes.
With Kiwacht-Edge, your communications remain in your control. SLMs assist, but never learn from you.
This transparency builds trust which opaque, cloud-based LLMs cannot match.
× CloseYour domain expertise is too valuable to train someone else’s model.
Why feed an impersonal foundry with your knowledge? Share what you know with your friends around a campfire.
× CloseKiwacht-Edge embodies ephemeral intelligence: it helps in the moment, then forgets.
No persistent storage means no breach risk, no audit nightmares, no hidden data lakes.
In an age of surveillance, forgetting is a feature.
× CloseAutomation without oversight leads to error and erosion of accountability.
Kiwacht-Edge is designed as a collaborative tool, it suggests, summarises, routes, but never acts without human approval.
This ensures ethical alignment, contextual accuracy, and user sovereignty.
× CloseWhether it's a Berlin tax office triaging citizen emails, a Lidl store monitoring oven alerts, or a creative agency managing client feedback - Kiwacht-Edge adapts.
Its workflow-agnostic architecture means you don't need separate AI tools for each channel.
One sovereign system. Infinite use cases.
× CloseMost AI optimises for engagement, data extraction, or scale.
Kiwacht-Edge optimises for human dignity, data sovereignty, and European values.
We believe Europe can lead in AI not through scale, but through sovereignty and by building intelligence which is non-extractive, but incorporates user-defined rules.
× CloseClick "more" on any of the cards above to see detailed explanations of our sovereign AI principles.