Commercial terms
Pricing
Indicative tariffs for business customers. Every price on this page is a projection prepared for the Founder's decision: it becomes a fact only with the first signed contract. Prices exclude VAT.
Dedicated instances
The first product is the dedicated instance — one customer per node, in an EU data centre, under a written agreement. Shared capacity (prepaid packs and pay-as-you-go) follows in a second phase. No plan renews without notice, no plan starts without an order, and cancelling is one action.
Every price: indicative — PROJECTED. No customer has paid yet; prices become established only with the first signed contract and the Founder's decision on live billing.
Dedicated CPU Instance
€499per month · excl. VAT · indicative — PROJECTED
PlannedA CPU-only node for the small model: classification, extraction and short answers at about 15 tokens per second per stream.
- Hardware: Hetzner AX42 — 8-core CPU, 64 GB RAM, NVMe (Germany/Finland)
- Models:
mh-sovereign-small - Projected capacity: ≈ 15 tokens/s per stream (MEASURED on the home CPU 2026-10-12; cloud parity PROJECTED), 2 concurrent streams, ≈ 16.6 M completion tokens per month at 35 % utilisation
- Availability: PLANNED — provisioned after a Founder spend GO; stated lead time up to 14 days
Lead product
Dedicated Instance 24 GB
€799per month · excl. VAT · indicative — PROJECTED
PlannedThe lead product: a 24 GB GPU node that holds all three served models, sized for teams that move from evaluation to production.
- Hardware: Hetzner GEX45 — NVIDIA RTX PRO 4000 Blackwell 24 GB GDDR7 ECC, 64 GB RAM, 2×512 GB NVMe (Helsinki, Finland)
- Models:
mh-sovereign-small,mh-sovereign-reason,mh-sovereign-lingua - Projected capacity: ≈ 24.6 tokens/s per stream (MEASURED on an RTX 5070 2026-10-12; same memory bandwidth — cloud figure PROJECTED), 8 concurrent streams, ≈ 81.5 M completion tokens per month at 35 % utilisation, ≈ 103,000 requests per month
- Availability: PLANNED — provisioned after a Founder spend GO; stated lead time up to 14 days (provider stock)
Dedicated Instance 48 GB
€2,490per month · excl. VAT · indicative — PROJECTED
PlannedA 48 GB GPU node for higher throughput and, on request, your own open-weight model up to about 30B parameters at 4-bit.
- Hardware: OVHcloud l40s-90 — NVIDIA L40S 48 GB, 15 vCore, 90 GB RAM (France)
- Models:
mh-sovereign-small,mh-sovereign-reason,mh-sovereign-lingua, customer-supplied open-weight model ≤ 30B 4-bit (on request) - Projected capacity: ≈ 31.6 tokens/s per stream, 8 concurrent streams (32 with the batching engine — PLANNED), ≈ 105 M completion tokens per month at 35 % — PROJECTED
- Availability: PLANNED
Dedicated Instance 80 GB
€4,490per month · excl. VAT · indicative — PROJECTED
PlannedAn 80 GB GPU node for the heaviest workloads and, on request, your own open-weight model up to about 70B parameters at 4-bit.
- Hardware: OVHcloud h100-380 — NVIDIA H100 80 GB, 30 vCore, 380 GB RAM (France)
- Models:
mh-sovereign-small,mh-sovereign-reason,mh-sovereign-lingua, customer-supplied open-weight model ≤ 70B 4-bit (on request) - Projected capacity: ≈ 73 tokens/s per stream, ≈ 242 M completion tokens per month at 35 % — PROJECTED
- Availability: PLANNED (quote)
Status labels — MEASURED — exists and was measured on the stated date by the stated method. BUILT — exists and works, not yet measured under load. PLANNED — designed, not yet built. A line without a label is not a claim of this platform.
Hardware, capacity and availability are quoted from the pricing source in English. Business customers only (B2B).
Included in every instance
- EU-owned infrastructure in an EU jurisdiction (Hetzner — Germany/Finland; OVHcloud — France)
- one customer per node (dedicated)
- zero retention of request content; audit metadata only (request id, model, status, token counts, latency)
- open-weight models with recorded licences (Qwen3.5 Apache-2.0, Gemma 4 Apache-2.0)
- OpenAI-compatible API format; streaming; tools; response_format
- API keys with visible prefix and secret stored only as a keyed one-way hash (HMAC-SHA-256); rate limits and monthly budget set by the customer
- metering ledger and GET /v1/usage; cabinet; status page
- Data Processing Agreement (Art. 28 GDPR) with sub-processor list — DRAFT, lawyer review pending (site /legal/dpa/)
- 2 hours of support per month
- availability target 99.5 % — PLANNED (target, not yet measured; SLA DRAFT at /legal/sla/)
- monthly billing; cancellation in one action in the billing portal
Add-on
Managed Execution (Sovereign AI Machine)
Planned€249per instance per month · excl. VAT · indicative — PROJECTED
Managed execution through Sovereign AI Machine on your instance: routed, multi-step work with honest busy and unavailable states and no external AI supplier, ever.
- the customer's instance driven by the project's Machine copy: POST /v1/respond with system, message, history; execution routes
- honest states (backend busy, upstream unavailable) mapped to the public error contract; no external AI supplier, ever
- tenant, project and key-prefix written to the audit line; request content never written
Availability: BUILT locally (Machine copy, tests) — cloud PROJECTED
Enterprise
Enterprise
Plannedfrom €3,990per month · excl. VAT · indicative — PROJECTED
Several dedicated instances with Managed Execution, a private network path, a custom data-processing annex, premium support and an annual contract with monthly invoicing.
≥ 2 dedicated instances, Managed Execution on all, private network path, custom DPA annex, premium support (8 h/month, 4-hour business-hours response), quarterly review, annual contract with monthly invoicing, optional customer-supplied open-weight model
Availability: PLANNED (quote-based; SLA and DPA exist as DRAFTS pending lawyer review)
Request an instanceSecond phase — shared capacity
Prepaid packs and pay-as-you-go need a shared pool, which is published only after the batching runtime is measured. Shown here so that you can plan; not on sale.
Prepaid packs
Planned| Pack | Price | Credit |
|---|---|---|
| Starter | €500 | €500 |
| Team | €2,000 | €2,100 |
| Scale | €10,000 | €11,000 |
Rules: 12-month validity; no auto-renew; balance visible in the cabinet; remaining balance refundable on request minus payment fees (Founder decision pending) · indicative — PROJECTED
Pay-as-you-go
Planned| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
mh-sovereign-small | €0.25 | €1.20 |
mh-sovereign-reason | €0.40 | €1.80 |
mh-sovereign-lingua | €0.60 | €2.60 |
PLANNED (phase 2) — published only after the batching engine (vLLM/SGLang) is measured; on llama.cpp the pool is not profitable · indicative — PROJECTED
How the prices are built
Each price stands above its cost floor at a 40 % gross margin; the floors come from the infrastructure quotes and the financial model of the platform.
| Plan | Price | Floor at 40 % margin |
|---|---|---|
| Dedicated CPU Instance | €499 | €446.94 |
| Dedicated Instance 24 GB | €799 | €668.97 |
| Dedicated Instance 48 GB | €2,490 | €2,012.54 |
| Dedicated Instance 80 GB | €4,490 | €3,625 |
| Managed Execution (Sovereign AI Machine) | €249 | €130.19 |
| Enterprise | from €3,990 | €3,243.08 |
Formula: price ≥ cost / (1 − margin) · indicative — PROJECTED
Payment and terms
Payments
PlannedCard or invoice through Stripe; we never store card numbers. Until the Founder authorises live payments the platform runs in Stripe test mode — no real charge is possible, and no plan is sold.
- **No hidden subscriptions.** A dedicated instance is billed monthly and renews only for as long as you keep it; cancellation is one action in the billing portal, effective at the end of the paid period.
- **Prepaid credit is valid for 12 months** and never renews by itself; the rule for refunding a remaining balance is awaiting the Founder's decision.
- **Business customers only.** Company name, registration number and VAT number are asked for before a paid service is activated; consumers are not served.
VAT
Prices exclude VAT. Estonian VAT of 24 % applies to customers in Estonia. Business customers in other EU member states with a valid VAT identification number are invoiced under the reverse-charge mechanism, with both VAT numbers on the invoice. Business customers outside the EU are invoiced outside the scope of EU VAT.