AGENTYK

European Sovereign AI Infrastructure

From Edge to Cloud

by Sylvanity B.V. — Delft, The Netherlands

Scroll

Architecture

The Agentyk Ecosystem

A complete sovereign AI stack — from infrastructure to applications, all EU-hosted

APPLICATIONS
PLATFORMS
AgentykCloudAgentykScribeAgentykAPI
ENGINES
InferenceForecastingRAG
LANGUAGE MODELS
HummingbirdMongooseGazelleGreyhoundCheetahLeopardJaguarTigerGrizzlyOrca
INFRASTRUCTURE
CloudEdge DevicesOn-Premise

Hybrid: apps automatically switch between edge and cloud based on task complexity

Language Models

AgentykLM — 10 Tiers

Named after nature’s fastest and strongest. Pick the tier that fits your workload — from instant on-device replies to frontier multimodal reasoning, with context windows up to 256K.

1

Hummingbird

Text

On-device, mobile, instant text completion

2

Mongoose

Text

Classification, intent detection, routing

3

Gazelle

Multimodal

Conversation, chatbots, low latency

4

Greyhound

Text

High-volume API, data extraction

5

Cheetah

Multimodal

Real-time translation, live support

6

Leopard

Multimodal

Long-context document processing

7

Jaguar

Multimodal

Complex reasoning, coding, math

8

Tiger

Text

Enterprise reasoning, multi-step logic

9

Grizzly

Multimodal

Multi-agent, scientific workloads

10

Orca

Text

Frontier: the most complex problems

Naming: AgentykLM [Animal] v[Version] — e.g. AgentykLM Jaguar v1

Where each tier lands

Pick the smallest tier that clears the bar your workload needs. Each tier is placed by the metric that is actually published for it — no marketing numbers.

Human preference

Relative standing only, no absolute scores. The top tiers, shown against well-known references.

AgentykLMExternal reference
  1. GPT-5.5-high
  2. Gemini 3.1 Pro
  3. Claude Opus 4.7
  4. AgentykLM Orca 1.0
  5. AgentykLM Tiger 1.0
  6. AgentykLM Grizzly 1.0
  7. GPT-5-high
  8. AgentykLM Jaguar 1.0
  9. Mistral Large 3
  10. Claude Opus 4.1
  11. GPT-OSS-120B
  12. Llama 3 70B

Capability

Relative standing only, no absolute scores. The remaining tiers, shown against leading open references.

AgentykLMExternal reference
  1. Qwen3.6 27B
  2. AgentykLM Greyhound 1.0
  3. EXAONE 4.5 33B
  4. AgentykLM Leopard 1.0
  5. Nemotron Cascade 2 30B
  6. gpt-oss-20B
  7. AgentykLM Cheetah 1.0 *
  8. AgentykLM Gazelle 1.0
  9. Ministral 3 8B
  10. AgentykLM Mongoose 1.0
  11. Phi-4
  12. AgentykLM Hummingbird 1.0 *
  13. Llama 3.2 1B

* closest-variant estimate — that exact build isn’t publicly rated yet.

Engines

Pure C runtime for Local Inference

Runs AgentykLM models on any hardware — Raspberry Pi to NVIDIA data centers.

Pure CLanguage

No C++, no external dependencies

UniversalGPU

Vulkan, Metal, CUDA

OptimizedCPU

NEON (ARM), AVX2, AVX-512

SmallestBinary

In the industry

FastestSpeed

Highest throughput per watt

ZeroDeps

Fully self-contained

Same DNA across all three: lower energy costs, smaller ecological footprint. Ideal for edge deployment.

agentyk-engine
$ agentyk-engine --build
Compiling inference_engine.c ...
Linking binary (424 KB) ...
Target: aarch64-linux-gnu
GPU backend: vulkan (auto-detected)
✓ Build complete — zero dependencies
$ agentyk-engine --bench gazelle-nederland-v1
Loading model ... 2.1 GB (Q4_K_M)
Warmup: 50 tokens @ 847 tok/s
Benchmark: 1 000 tokens @ 892 tok/s
Memory: 2.3 GB peak | Power: 12.4 W
✓ 72.1 tokens / sec / watt

Products

The Agentyk Suite

Web, desktop, mobile, CLI — sovereign AI on every surface. All apps switch seamlessly between edge and cloud inference.

AgentykCloud

Cloud Platform

Privacy-first AI chat. Stateless server, conversations live only in your browser, broad model selection. 100% EU-hosted.

  • Stateless server — no database
  • Conversations stay in your browser
  • AgentykLM + open models
  • 100% Dutch hosted
Learn more →

AgentykAPI

RAG-as-a-Service

Enterprise knowledge management with Retrieval-Augmented Generation. EU regulations, legislation, scientific publications.

  • Auto-indexed vector DBs
  • Source-cited answers
  • Custom knowledge bases
  • Daily update cycles

AgentykChat

Desktop Client

Local inference on Windows, Mac, and Linux. Privacy-sensitive tasks run fully offline with seamless cloud switching.

  • Full offline capability
  • Local inference engine
  • Seamless cloud fallback
  • Office-ready
Learn more →

AgentykMobile

iOS & Android

On-device AI for field workers, technicians, and sales teams. Runs Hummingbird to Gazelle models without internet.

  • On-device models
  • No internet required
  • Field-ready UX
  • Edge-first design
Learn more →

AgentykCode

CLI Tool

Command-line interface for code generation, review, and debugging. Integrates with existing IDEs, local + cloud models.

  • Code generation & review
  • IDE integration
  • Local Code models
  • Cloud fallback
Learn more →

European Sovereignty

Privacy by Design

EU-sovereign AI free from the CLOUD Act. Your data stays in the EU/EEA — always.

Stateless Server

Server-side has no database — there's nothing for us to lose, leak, or hand over

Stays Local

Conversations live in your browser, never on our servers

EU

EU-Hosted

Data stays in the EU/EEA

GDPR

Compliant by design, EU-only processing

US-based AI vs. Agentyk

AspectUS-based AIAgentyk
CLOUD ActUS can seize data, even in EUNot applicable
Data trainingPrompts may train modelsNo data used for training
GDPRLegal grey areaCompliant by design
Data locationUS or unclearEU/EEA-hosted
PersistenceConversations stored server-sideBrowser-only, never our servers

Read the full Privacy Policy →

Responsible Compute

Sustainability by Design

Ten model tiers let you run only the compute a task needs. And our servers use no water to stay cool.

10

10 Tiers

Ten model sizes, so there is always a right-sized option and never a giant model for a small job

Right-Sized

Pick the smallest tier that handles your prompt; complex work gets the large models, simple work does not

0L

Zero Water

Our servers use no water coolant. Nothing is withdrawn or evaporated to keep them cool

Lower Energy

Smaller tiers draw far less power per answer, so matching size to task cuts energy use directly

Typical AI cloud vs. Agentyk

AspectTypical AI cloudAgentyk
Model selectionOne large model for every requestTen tiers, choose the size the prompt needs
Energy per requestHeavy, even for trivial promptsScales down with the tier you pick
CoolingEvaporative water cooling, high water drawNo water coolant, zero water use
EfficiencyOverprovisioned by defaultRight-sized by design

See the ten model tiers →

Pricing

Pick a plan

Flat monthly fee, auto-renewing. EU-billed via Mollie. Plus includes a 14-day free trial.

Cancel any time — with one click

Plus

For everyday AI use.

20/ month

incl. 21% VAT

14-day free trial

  • Generous monthly usage allowance
  • AgentykLM Cheetah today; Jaguar / Tiger / Grizzly / Orca rolling out
  • PDF / DOCX / TXT / MD upload up to 50 MB
  • In-process tools (calculator, units, doc Q&A)
  • Email support
Get started

Pro

Recommended

Power users and small teams.

100/ month

incl. 21% VAT

  • 5× the Plus usage limits
  • Same frontier-tier model lineup
  • SDK access for on-device inference
  • Priority email support
Get started

Pro Max

Heaviest individual workloads.

200/ month

incl. 21% VAT

  • 20× the Plus usage limits
  • Everything in Pro
  • Priority response queue
Get started

Enterprise

Teams, SLA, regulated workloads.

Let's talk
  • Unlimited usage
  • Multi-user, multi-device, SSO
  • GDPR / DPA paperwork, SLA
  • Air-gap and on-prem options
  • Dedicated support
Contact sales