Tier Raw
Compute rent — self-serve
€150 / month
160 hours included · €1.50/hr overage
- SSH access to our DGX Spark GPU
- OpenAI-compatible API endpoint
- Bring your own model & code
- Fair-use policy — predictable billing
AI for your business.
No GPU management. No
data leaks.
You bring the use case. We bring the models, the infrastructure, and the fine-tuning expertise. Three pricing models — pick the one that fits your business.
Your data never leaves your control. You pay how you want: monthly subscription, per compute minute, or as a depreciable investment.
Our offer
Three tiers, one promise
Every tier starts with a conversation. We scope the work, define success, and give you a fixed price. No surprises.
Tier Rent
Fine-tuned model — pay per use
€250–€800 / month
by model size · 160 hours included
- We QLoRA fine-tune on your data
- Deployed on our infrastructure
- Guaranteed machine availability
- Concurrent sessions by tier
Tier On Premise
Fine-tuned model — investment pricing
€5k–€20k initial
- We QLoRA fine-tune on your data
- Deployed on your infrastructure
- Pay as investment — depreciable asset
- Your data never leaves your network
Not sure which tier fits? Request a custom offer — we'll scope it together.
Tier Raw
Rent GPU compute, predictable monthly cost
SSH into our DGX Spark, run any model, pay a predictable monthly fee with a fair-use policy. No fine-tuning included — you bring your own code, your own model, your own data.
What you get
- Unix account on a dedicated DGX Spark (NVIDIA GB10, 128 GB unified memory)
- SSH access via Tailscale — secure, no static IP needed
-
OpenAI-compatible API endpoint
(
/v1/chat/completions,/v1/models) - 50 GB disk quota for models and data
- Unique API key per user
- 1 concurrent session (standard plan)
Pricing
| Component | Price |
|---|---|
| Monthly subscription | €150 EUR |
| Included compute | 160 hours / month |
| Additional hours | €1.50 EUR / hour |
| Minimum session | 1 minute |
| Billing | Monthly invoice, post-paid |
Fair-use policy
Your monthly subscription covers up to 160 hours of compute — typical 9-to-5, 5-days-a-week usage. If you need more, additional hours are billed at €1.50/hr with no cap. This ensures predictable costs for regular use while keeping the machine available for all tenants.
How it works
- Tell us you want access — we'll create your account
- You SSH in, run inference, or use the API
- When you're done, we compute the bill
- We send an invoice — you pay via bank transfer, PayPal, or Stripe
Tier Rent
Custom fine-tuned model, guaranteed availability
We fine-tune a model on your data using QLoRA, deploy it on our infrastructure, and you call the API. You pay a monthly subscription based on model size, with guaranteed machine availability and concurrent sessions.
What you get
- QLoRA fine-tune on your proprietary data (cleaned by us or we help)
- Fine-tuned LoRA adapter deployed on our inference server
- API endpoint — integrate into your CRM, support system, or internal tool
- Optional simple dashboard with usage stats
- Guaranteed inference capacity — your model always available during business hours
- We handle all ML engineering — you focus on your business
Model-size tiers
| Model size | Monthly | Included hours | Additional hour | Concurrent sessions |
|---|---|---|---|---|
| Small (≤13B params) | €250 | 160 hours | €0.50 | 1 |
| Medium (35B params) | €500 | 160 hours | €0.75 | Up to 2 |
| Large (70B params) | €800 | 160 hours | €1.00 | Up to 4 |
Pricing notes
- Discovery phase: fixed fee — see custom offer below. Credited if you proceed.
- Guaranteed availability: your model has reserved inference slots during business hours (9–5, weekdays). No queue contention.
- Concurrent sessions: higher tiers unlock parallel request handling via DGX Spark hardware concurrency. Actual throughput scaling depends on model and workload.
- Volume discounts: available for high-throughput customers — contact us.
Why SMEs prefer this
- No upfront: pay when the model works, not before
- Predictable: monthly subscription + transparent overage
- Guaranteed: your model gets dedicated capacity during peak hours
- Flexible: stop anytime, no lock-in
- Secure: your data stays with us, not on public API servers
Tier On Premise
Own your AI. Deploy on your infrastructure.
We fine-tune on your data, then deploy the model on your hardware or a dedicated instance we provision. You pay as an investment — capital expenditure, depreciable over 3–5 years. Tax advantages, full data control, predictable total cost of ownership.
What you get
- QLoRA fine-tune on your proprietary data
- Deployment on your GPU server or a dedicated DGX Spark (we provision)
- Model never leaves your network — full data privacy
- Ongoing support: retraining as your data grows
- Fixed-price proposal after paid discovery phase
Pricing
| Component | Price |
|---|---|
| Discovery phase | Fixed fee — see custom offer |
| Initial fine-tune + deployment | €5,000–€20,000 (one-time) |
| Annual maintenance / retrain | 15–20% of initial fee |
| Hardware (if we provision) | Pass-through + 10% margin |
Hardware options
- Your hardware: we deploy on your existing GPU server
- We provision: we set up a dedicated DGX Spark at your site or colo
- Hybrid: we manage, you own — predictable annual cost
Get started
Request a custom offer
Every engagement starts with a paid discovery phase. We scope your use case, define success metrics, and deliver a fixed-price proposal. If you proceed, the discovery fee is credited toward the engagement.
This is our commitment: we don't pitch one-size-fits-all. We listen, diagnose, and propose something specific to your business.
What the discovery covers
- Understand your use case and data
- Define success criteria (accuracy, latency, throughput)
- Scope the fine-tune (base model, QLoRA config, duration)
- Deliver a fixed-price proposal with timeline
How to start
Send us an email with a brief description of your use case, approximate data volume, and which tier interests you. We'll respond within 48 hours with a discovery offer.