SMT 6.1 — Private AI Power for enterprises that demand control, performance, and scale.
A scalable enterprise AI infrastructure for private cloud, agentic workflows, long-context processing, and high-performance LLM applications — running on dedicated GPU power instead of limited standard subscriptions.
Individual technical assessment per our terms
Swiss Modern Technologies.
SMT is the technology arm of Swiss Marketer. Our conviction: Swiss businesses shouldn't have to rent modern AI forever like someone else's subscription — they should run and control it as their own infrastructure. That's what we're building: step by step, tool by tool, with Swiss standards for precision, data protection and reliability. SMT 6.1 is the current state of that work — not the end of it.
Standard AI is powerful. But not enough for every enterprise.
Limited control
Infrastructure, model behavior, and data flows sit with the provider — not with you.
Limited scale
Token limits, rate limits, and shared capacity slow you down exactly when load peaks.
Limited integration
CRM, telephony, databases, and internal systems stay out of reach — AI works in isolation.
Limited data privacy
Sensitive company data in shared environments — location and options often not yours to choose.
Private Cloud instead of a standard subscription.
- Access: protected workspace with access controls, roles and logging.
- Transport: encrypted end to end (TLS); private data rooms per customer.
- Go-live: every instance passes a security audit before handover.
- Training: your data is never used for model training — contractually guaranteed (details in the T&C).
Advanced Protection is project-dependent and available on request at additional cost.
Your own GPU instances instead of shared capacity — performance reserved for you.
Data location by project requirement — depending on availability and region.
Your company knowledge as a private, searchable data space.
Agents and automations tailored precisely to your processes.
Cost and capacity follow demand — not artificial usage limits.
More transparency into where data flows and how it is processed.
Connections to CRM, website, databases, and telephony available.
Extended protection measures as an optional, project-dependent module.
Operating models — from dedicated instance to AI factory
SMT 6.1 can be deployed across a wide range of operating models: from a single dedicated instance to multi-region setups, air-gapped private cloud or hybrid environments. With load balancing, auto-scaling, backup, disaster recovery and dedicated SLAs.
A dedicated GPU instance for one tenant — simple, isolated and fast to provision.
Distribution across multiple regions — including multiple EU locations where available.
Fully isolated infrastructure for maximum control and data privacy.
Combinations of on-premise and cloud tailored to your compliance requirements.
Automatic load distribution and scaling based on workload and request volume.
Regular backups, disaster recovery and dedicated service-level agreements. Per our terms, SLAs apply only if expressly agreed in writing.
One infrastructure. Many business processes.
Score leads, prepare follow-ups, prioritize pipeline actions.
Understand inquiries, draft responses, structure and escalate cases.
Produce campaign assets, variants, and analyses at scale.
Read, extract, compare, and summarize large document collections.
Transcribe calls, analyze them, and connect them to actions.
Company knowledge as a private assistant — answers from your own data spaces.
Structure receipts, prepare reports, streamline administrative processes.
Create property exposés, analyses, and advisory documents from data.
// SMT 6.1 workspace — signed in · instance: dedicated · region: EU You: Analyse the 12 attached contracts and flag every deviation from our standard framework agreement. SMT 6.1: Processing 12 documents in parallel … 4 deviations found — details with citations: …
All code and configuration examples are illustrative and may differ from the final product.
Scale AI agents in real time — with control.
With SMT 6.1, multiple AI agents work in parallel: analyzing, writing, reviewing, structuring, classifying, prioritizing, responding, and preparing actions. SMT 6.1 enables agent workflows that scale in parallel for complex enterprise tasks — with flexible Multi-GPU compute for peak loads.
How many agents can work concurrently depends on hardware, model configuration, data structure, workload and response-time targets, and is sized per project. *Details per our terms and individual technical assessment.
SMT 6.1 orchestrates parallel specialized AI agents — each with a defined scope, fixed guardrails and role-based tool access. High-impact actions trigger a human-in-the-loop approval; every step is recorded in an audit trail. Budget and concurrency limits prevent cost and load spikes.
Guardrails & roles
Each agent may only use its approved tools and data — defined by role.
Human-in-the-loop
Critical actions are submitted for approval before execution — never without control.
Audit trail
All agent actions, tool calls and decisions are logged in a traceable way.
Budget & limits
Cost and concurrency quotas per team or agent protect against overload.
Requested today — productive in days.
No months-long infrastructure project: we provision your dedicated SMT 6.1 instance, harden it and hand over protected access — you get going right away.
Sizing call
Use case, data landscape, security requirements — from that we define GPU tier, operating mode and region (EU).
Provisioning & hardening
Dedicated instance, TLS, access controls, security audit — fully done for you. No DIY kit, no ops team required.
Launch & scale
Protected workspace from day one. More GPUs, more agents, more context — bookable any time, without migration.
*Depending on GPU availability, region and project scope.
How you work with your instance.
We set up your private instance and hand over the access keys. From then on your systems talk directly to your AI — through an interface that follows the OpenAI standard. Which means: existing programs and tools usually work without modification; only the address and the key change.
Interface for your systems
Your software queries the AI directly — streaming answers as they are written, custom function calling, answers in a fixed data format, and a mode where the AI works through several steps on its own.
Overview in the customer portal
Your portal shows the state of your instance: whether it is running, which security level is set, and how many requests per minute are available.
Keys & limits
We create the access keys, grant each key only the permissions it needs, set daily quotas and can revoke any key instantly. Every key is bound exclusively to your instance.
Setup, key management and permissions run as a managed service through Swiss Marketer — deliberately no self-service, so access stays traceable. A desktop application (SMT Studio) is currently used internally and is not yet part of the customer offering.
From a single GPU machine to a scalable AI Factory.
H100 SXM
For high-performance LLM applications, internal assistants, automations, and production workloads.
H100 NVL
For larger models, stronger inference performance, and more demanding multi-user systems.
H200
For memory-intensive workloads, long contexts, larger models, and data-intensive applications.
B200
For very high parallelization, multi-agent systems, large enterprise models, and complex workloads.
B300
For maximum enterprise scale, agentic parallel processing, large context architectures, and performance-intensive AI factories.
Availability, location, pricing, and specific performance figures may vary by provider, region, capacity, and project requirements.
Multiple GPU machines can be combined to cover parallel agents, distributed inference, RAG pipelines, high user load, and on-demand compute.
Cost based on compute — not artificial usage limits.
Inference Setup
- Internal assistants
- Chatbots
- Simple automations
- Production LLM applications
Agentic Business Setup
- Multiple agents & workflows
- Knowledge bases
- RAG systems
- Operational automations
Enterprise AI Factory
- GPU clusters & multiple machines
- High user load
- Agentic parallel processing
- Dedicated protection architecture
Pricing on request. EU hosting selectable depending on location and availability. Advanced Protection on request at additional cost.
Price on requestSetup + compute — cost and capacity follow demand.
Request setupSMT 6.1 in detail — the architecture behind the performance.
SMT 6.1 is a latest-generation large language model, agentically optimised and operated as private infrastructure — with a hybrid reasoning architecture, efficient expert routing and two switchable operating modes.
Answer directly
The AI answers without a preceding thinking step — right for assistants, automations and anything that should come back fast.
Think first, then answer
The AI works through the task visibly in several steps before answering — for complex analyses, large document sets and multi-step agent chains. Switchable on every single request.
*Orientation values — depending on hardware, model configuration, workload and availability. Details per T&C and individual technical assessment.
- Model class
- Frontier-class LLM of the latest generation — for analysis, code, agents and automation.
- Architecture
- Mixture-of-Experts transformer — hundreds of billions of parameters in total, only a fraction active per token: frontier performance with efficient inference.
- Reasoning
- Thinking on demand: per request, choose between a direct answer and visible multi-step thinking for complex analysis.
- Context
- Around 262,000 tokens per conversation* with four concurrent conversations — correspondingly more for single very long conversations. Extendable beyond that via vector index, memory layer and chunking.
- Agents
- Native tool calling, multi-step planning, self-correction — parallel scalable agent workflows*.
- Languages
- German, English, French, Italian and more — plus code (SQL, Python, TypeScript, …).
- Output
- Text, structured data (JSON), function calls, automations — straight into your systems.
- Hardware
- Dedicated NVIDIA GPU instances — H100 to B300, single machine or cluster, EU hosting selectable.
- Privacy
- Your data is never used for model training. Encrypted transport, access only after a security audit.
- API
- OpenAI-compatible REST API on your instance — streaming chat, custom function calling, fixed-format (JSON schema) answers, agent mode.
A lot of text at once — and beyond.
An AI model can only keep as much in view as fits its context window — put simply: its working memory. In the current sizing that is around 262,000 tokens (roughly, word fragments) per conversation, with four conversations running concurrently. Size an instance for a single very long conversation and considerably more is possible — both at once is not achievable on the same machine. For data sets beyond even that, a vector index, memory layer and chunking take over.
Context length, speed, and quality depend on model configuration, hardware setup, data structure, workload, and availability. Details per our terms, privacy policy, and individual technical assessment.
Commercial AI models vs. SMT 6.1 Private Infrastructure
The leading commercial models are outstanding — each with its own profile. The difference is not about “better or worse,” but about the operating model: a shared SaaS platform or your own private infrastructure.
| Factor | Commercial AI subscriptionsOpenAI · Gemini · Claude · Perplexity | SMT 6.1Private Infrastructure |
|---|---|---|
| Operating model | External SaaS / API — you rent access to a shared platform. The provider sets pricing, limits, and features | Private cloud or dedicated GPU infrastructure — you operate your own AI layer. Hardware, location, and setup selectable per project |
| General model quality | Frontier models set the standard for general model quality and breadth | Specialized and configurable to your processes and data — quality depends on setup and configuration |
| Data control & location | Processing and location as defined by provider, contract, and region | Controllable per project — EU hosting selectable, private data spaces |
| Scaling under load | Rate limits, plan tiers, and shared capacity — peak load as defined by the provider | Elastic multi-GPU performance — scaling with compute demand, from a single machine to a cluster |
| Agent parallelism | Parallelism possible via API, bounded by the provider’s limits and cost logic | Parallel-scalable agent workflows — up to hundreds of agents, depending on infrastructure and workload* |
| Cost logic & usage limits | Per token, per user, or per subscription — ongoing, usage-based. Token, rate, and fair-use limits per plan | Based on setup, hardware, location, and the compute you use. No artificial usage limits — the boundary is the compute you book* |
| Integration depth & customizability | Via the provider’s official interfaces and connectors. Prompts, in part projects/GPTs and API features — within the platform’s scope | Direct integration with CRM, website, databases, and telephony available. Prompts, RAG, private knowledge bases, your own workflows, and optional fine-tuning |
| Security options | Enterprise options as defined by provider and contract tier | Access controls, encryption, optional data isolation — Advanced Protection on request |
All providers side by side
| Criterion | SMT 6.1 | OpenAI · GPT | Google · Gemini | Anthropic · Claude | Meta · Llama | Mistral | DeepSeek | Perplexity |
|---|---|---|---|---|---|---|---|---|
| Dedicated private instance, done-for-you | ✓ | partial¹ | partial¹ | partial¹ | partial² | partial² | partial² | — |
| Selectable data location (EU) | ✓ | partial | partial | partial | partial² | partial | partial² | — |
| Contractually no training on your data | ✓ | partial | partial | partial | partial² | partial | partial² | partial |
| Agentic context architecture (RAG/memory) included | ✓ | separate | separate | separate | separate | separate | separate | — |
| Speed / Deep operating modes on your own hardware | ✓ | — | — | — | partial² | partial² | partial² | — |
| Agent parallelism without provider rate limits | ✓* | — | — | — | partial² | partial² | partial² | — |
| Costs by compute instead of token metering | ✓ | — | — | — | partial² | partial² | partial² | — |
| Customisation on company data incl. operations (RAG/fine-tuning) | ✓ | partial | partial | partial | partial² | partial | partial² | — |
| Integrated business platform (CRM · telephony · website) | ✓ | — | — | — | — | — | — | — |
| Security audit before go-live included | ✓ | — | — | — | — | — | — | — |
| Public API | ✓ (on your instance) | ✓ | ✓ | ✓ | partial³ | ✓ | ✓ | ✓ |
| Web search with source grounding | partial (RAG + optional) | partial | partial | partial | — | partial | partial | ✓ |
Legend: ✓ = part of the offering · partial = possible with restrictions or extra effort · separate = to be built as its own project · — = not core to the offering.
¹ via the provider's enterprise/cloud programmes · ² self-host / open weights, self-operated — operations, hardening and scaling on you · ³ via third-party hosting. Classification by typical, publicly communicated operating models (as of 2026), no claim to completeness; no statement about model quality.
Fairly positioned: The commercial frontier models are the right choice for many tasks and set the standard for general model quality. SMT 6.1 does not compete as a “better model” — but as private, controllable infrastructure for enterprises where data control, location, integration depth, and scaling by compute are decisive. The two can be combined.
*Reference values — depending on hardware, model configuration, workload, and availability; details as per our terms and an individual technical review. Statements about third-party providers describe typical operating models without any claim to completeness; conditions as per the respective provider. All brand and product names are the property of their respective owners; no partnership or endorsement is implied.
The SMT 6.1 API — your AI, programmable.
An OpenAI-compatible REST API on your dedicated instance: same SDKs, same patterns — but your private infrastructure behind it. We set up your instance and hand over the access keys.
curl https://portal.swiss-marketer.ch/api/smt/v1/i/YOUR-INSTANCE/chat/completions \ -H "Authorization: Bearer $SMT_API_KEY" \ -d '{ "model": "smt-6.0", "messages": [{ "role": "user", "content": "Analyse the Q3 report …" }], "stream": true }'
All code and configuration examples are illustrative and may differ from the final product.
Connect SMT 6.1 to your stack via an OpenAI-compatible API, webhooks and native connectors. Through Make, Zapier or n8n, 2000+ tools are available — complemented by direct integrations with ERP, CRM, DMS, BI and identity providers.
Real-time triggers for workflows, notifications and automations.
Direct integration with ERP, CRM, DMS, BI and other enterprise systems.
Connect 2000+ tools without coding — ideal for citizen integrators.
Roles, access and provisioning centrally through your identity provider.
Customers receive the full technical documentation during onboarding.
You're on the waitlist 🎉
We'll reach out with your private SMT 6.1 API access as soon as your instance is ready.
AI as your own infrastructure — not someone else's subscription.
In a conversation, we clarify your use case, data situation, security requirements, and the right GPU tier — and define a setup that scales with your business.
Performance figures, context lengths, parallelization, availability, privacy options, and protection measures depend on hardware, location, model configuration, workload, data structure, and contract scope. Details per our terms, privacy policy, and individual technical assessment.