Services

Four services. One accountable partner.

Each service can be bought on its own, but they are designed to fit together: the same engineer who writes the design document is the one who gets called at three in the morning.

Dedicated inference

A reproducible serving stack for open-weight models, provisioned on capacity that is not shared with anyone else.

  • Capacity planning from your prompts and concurrency, not from vendor benchmarks
  • vLLM or SGLang, quantisation chosen against a measured quality threshold
  • Autoscaling within a hard ceiling, so a runaway client cannot empty your budget
  • Load and soak testing before go-live, with the numbers written down
  • Blue/green model rollouts with a tested rollback path

Private retrieval

Grounded answers over your own document estate, without turning your permissions model into a suggestion.

  • Connectors for the systems you already run: file shares, wiki, ticketing, CRM
  • Access control resolved at query time from the source system
  • Hybrid keyword and vector search, with a re-ranking stage where it earns its latency
  • Citation back to the source document, so an answer can be checked
  • Ingestion pipeline with delete propagation when a document is removed

Identity, governance and audit

The controls your security team will ask about, implemented rather than described.

  • SAML 2.0 or OIDC single sign-on, with group-to-role mapping
  • Per-team and per-application token budgets, enforced at the gateway
  • Prompt and response retention you define, from zero logging to full archive
  • Request-level audit export to Splunk, Elastic or your SIEM of choice
  • Policy templates for acceptable use, data classification and incident handling

Managed operations

We keep the platform patched, monitored and inside budget after the project team has moved on.

  • Monitoring, alerting and an on-call rotation with a named engineer
  • Driver, runtime and CVE patching on a published cadence
  • Model upgrades gated by an evaluation run on your own test set
  • Monthly service report: incidents, latency, spend, capacity headroom
  • Quarterly review with your platform and finance stakeholders
Engagement models

Published ranges, quoted in writing.

Ranges below are typical for a single business unit. Every engagement is quoted after scoping, and infrastructure is always billed at cost.

Assessment

Feasibility and design

EUR 6,000 - 9,000 · 1-2 weeks

A written answer to one question: can this workload run privately, on what hardware, at what cost, and what would break first?

  • Workload and data-flow review
  • Reference architecture
  • Capacity and cost model
  • Risk register and go/no-go recommendation

Build

Deployment to production

EUR 28,000 - 60,000 · 3-6 weeks

The platform, built in your environment, hardened, integrated and handed over with documentation your team can actually use.

  • Inference and retrieval stack
  • SSO, quotas and audit pipeline
  • Load test and acceptance run
  • Runbooks and a handover workshop

Operate

Managed service

From EUR 4,500 / month

Ongoing responsibility for availability, patching, model lifecycle and cost control, with a monthly report you can forward to finance.

  • 24/7 monitoring and on-call
  • Patch and upgrade management
  • Monthly service report
  • Quarterly capacity review
What is not included

The parts we deliberately do not sell.

  • No licence resale and no per-token markup. You buy capacity directly at cost.
  • No proprietary runtime that only we can operate. Everything ships with documentation and root access stays with you.
  • No application development. We run the platform your product sits on, not the product itself.
  • No training on your data. Fine-tuning is a separate engagement with its own contract, and only if you ask for it.
How an engagement starts

A short email is enough.

  • You send the workload, the data classification and the deadline.
  • We reply within one business day with questions, or with a reason we are not a fit.
  • A 45-minute call with an engineer, not a sales deck.
  • A written scope and fixed price before anything is signed.

Not sure which of the four you need?

Most engagements start with the assessment. If your workload is small enough to run on a single node, we will say so instead of selling you a platform.