+ vostro [ital.] — yours
VOTRO deploys frontier open-weight language models on infrastructure you own or control. Nothing leaves. The model is yours — weights, context, keys, costs.
$ votro init --site your-lan --mode air-gap
→ weights: glm-4.6 · verified sha256 ✓
→ egress: 0 bytes · retention: 0 days
→ tenancy: single · root: yours
→ status: CONTAINED
$
+ Statement of purpose
Public AI asks you to trust someone else’s cloud with your most sensitive work. We think that trust is misplaced — not because clouds are careless, but because it should never be required. VOTRO installs the model itself on hardware you control, turning confidentiality from a policy into a physical property of your infrastructure.
Contracts can be breached, certifications can lapse, interfaces can change. Physics cannot be argued with. What is yours stays yours.
+ Built for
+ 01 — Products
On-premise
Frontier models installed inside your own data centre or office. Nothing crosses your firewall — not prompts, not weights, not logs.
Dedicated hosted
Dedicated GPU infrastructure locked to your keys, region of your choice. We retain and manage the hardware — one monthly fee covers lease, licences and operation. You keep root.
+ 02 — Model catalogue
| MODEL | ROLE | SIZE / CONTEXT | NOTES |
|---|---|---|---|
| GLMFLAGSHIP | Flagship | 355B MoE · 200K context · MIT | Default frontier deployment. Leads open-weight agentic benchmarks — tool use, long-horizon coding, document analysis. |
| DeepSeek | Reasoning | 671B MoE · 128K context · MIT | Frontier-grade reasoning and mathematics at a fraction of frontier cost. |
| Qwen | Generalist | 0.6B–235B · 128K context · Apache 2.0 | A model for every slot, from edge-scale to frontier MoE, with strong multilingual coverage. |
| Mistral | Fast | 24B dense · 128K context · Apache 2.0 | Dense, efficient, dependable. Serious utility on a single node. |
+ 03 — Process
We map your workflows, data sensitivity and hardware, then select models, sizes and guardrails — starting with a fixed-fee pilot.
Our engineers install, tune and integrate the stack behind your existing access controls. Typical go-live: two to six weeks.
Monitoring, updates and optimisation for as long as you run it — with full documentation and handover the day you ask for it.
+ 04 — FAQ
The best open-weight models available — GLM, DeepSeek, Qwen, Mistral — selected against your accuracy, latency and budget targets. You can run several side by side and swap as the frontier moves. You are never locked to one vendor's model.
No. Inference happens inside your perimeter, so prompts, documents and model responses never reach us. Where our engineers touch hardware at all, it is with your approval, under your supervision — and they still see none of your data.
A single multi-GPU server can serve a department; larger organisations run multi-node clusters. If you don't own hardware, we rent dedicated single-tenant GPUs in your name and your chosen region.
Yes. Fully offline deployments — including model and software updates delivered as signed offline bundles — are available for regulated and classified environments.
Those run on vendor clouds and rely on contracts and certifications. We move the model itself onto hardware you control, so confidentiality is a physical property of your infrastructure — and your costs are fixed, not metered per token.
ENCLAVE™ is one predictable monthly fee — typically $15k–40k depending on GPU capacity — covering the hardware lease, software licences and continuous management. No capex. VAULT™, on your own hardware, is a one-time setup and development fee (typically $250k–500k) followed by a $10k/month SLA covering maintenance, security patching and model upgrades as better open weights ship. Pilots are fixed-fee and credited against deployment.
A pilot typically goes live in two to six weeks: roughly one week to assess, one to three to deploy, and a parallel rollout with your team. Handover documentation is delivered from day one.
If your data can’t leave the building, your AI shouldn’t either. One call, fixed-fee pilot, handover from day one.