Your AI.
Your hardware.
Not a single byte leaves the building.
Quinta turns your own hardware into a private, OpenAI-compatible AI service — with enterprise management and a complete audit trail. Ready in hours, not months.
Pull the network cable. Quinta keeps answering — everything runs on your hardware.
Cloud AI means someone else’s infrastructure, someone else’s rules.
Data leaves Europe
The most sensitive company data goes to US cloud APIs today — without an audit trail, with GDPR, EU AI Act and NIS2 risk.
Costs without a ceiling
Token billing makes AI budgets unpredictable. Every new use case keeps pushing the running cost up.
Dependence on vendors
Models, prices and availability are set by an outside provider. Switching an API today means a rebuild, not a move.
Not a model runner.
The complete operating layer.
Every request takes the same path: through the gateway (the gatekeeper), to the daemon on your GPU machine, into the model — connected through a central registry. Every additional machine running the Quinta daemon becomes part of the platform automatically.
High-performance inference for the entire HuggingFace catalogue. Fully automated lifecycle: download → start → health check → warmup → ready.
The simple path to your first model — from a workstation to an AI appliance of the DGX Spark class. GPU detection for NVIDIA, AMD and Intel and automatic routing included.
Two lines.
That’s all it takes.
Existing applications move from OpenAI or Azure OpenAI by changing base_url and the API key. No reprogramming, no new SDK.
- /v1/chat/completions
- Embeddings
- Rerank
- Audio transcription
- Responses API
Business and Enterprise — for production use
Both editions run entirely on your own infrastructure. The difference is in scale, governance and support.
| Recommended Business On request License per site/instance | Enterprise Custom Custom contract & SLA | |
|---|---|---|
| Infrastructure | ||
| Quinta gateway & daemon | ||
| Swappable inference backends | ||
| OpenAI-compatible API | ||
| Compute nodes | Multiple | Unlimited |
| Multi-site deployment | ||
| Access & users | ||
| Organizations | Unlimited | Unlimited |
| API keys & scopes | ||
| Roles (RBAC) | ||
| Commercial use | ||
| Management | ||
| Model management | ||
| Usage analytics & metrics | ||
| Instance admin & invitations | ||
| SSO / SAML | ||
| 2FA & passkeys | ||
| Audit trail | ||
| Fine-tuning UI & MCP | ||
| Support | ||
| Support channel | Slack / Teams | |
| Response time | 24 h | Custom |
| Dedicated contact | ||
| Annual on-site check | ||
All pricing is based on self-hosting on your own hardware. Setup support and on-site operation are agreed individually in the offer.
Sovereignty costs no performance — it delivers it.
Benchmark of the underlying engine, measured on an NVIDIA DGX Spark (128 GB): 31,200 requests, up to 512 concurrent requests, Quinta gateway against vLLM alone. Identical per-user token rate.
Three steps to your own AI service.
Plan
One conversation: your use cases, your hardware, your compliance framework. The deployment plan follows.
Deploy
The Quinta daemon onto your GPU machines, gateway and dashboard alongside — nodes register themselves, models start automatically.
Operate
Deployments by click, API keys, usage per unit, Prometheus metrics and a complete audit trail in the dashboard.
Enterprise-grade, because it has to be
Banking and client confidentiality, trade secrets, GDPR, the Swiss DSG and the EU AI Act — Quinta is built for data and use cases that demand demonstrable governance.
Data sovereignty
Models run on your own hardware — in your data centre, not in someone else’s cloud.
GDPR by design
No data sharing with third parties, no training of foreign models with your data.
Roles & permissions
Fine-grained RBAC enforced server-side and client-side — not just hidden in the UI.
EU AI Act ready
A complete audit trail, traceable model deployments and human oversight meet the documentation and transparency duties for high-risk AI — where many cloud providers hit limits for lack of insight into their own infrastructure.
GDPR- & EU AI Act-compliant AI infrastructure
Until 2 August 2026 — high-risk obligations of the EU AI Act
Compliance
Compliance crunch
2 August 2026 is the date every CIO should know. The high-risk obligations of the EU AI Act apply in full from then. Now is the time to prepare your own AI stack.
- High-risk AI systems must be compliant by 2 August 2026
- Fines up to €15M or 3% of global annual turnover
- Your current cloud AI stack often does not meet the requirements
- Quinta supports compliance through architecture, not just paper
Legal
Legal position
Especially sensitive GDPR data should not leave the EU legal area. US cloud providers are subject to a different legal order — your data does not necessarily stay only yours there.
- Data never leaves EU legal jurisdiction
- No access risk through the US CLOUD Act
- Full control over processors
- Legal authority stays with your organization
Audit
Audit-ready?
Only demonstrably compliant AI infrastructure passes an audit. We help you reach that state before the deadline — not after.
- Complete audit trail for every inference request
- Model provenance and version logging
- Role-based access control built in
- Audit-ready from day one
Four login methods, freely combinable
Password
The standard login path.
2-factor (TOTP)
Numeric code from an authenticator app.
Passkeys
Sign in with fingerprint or Face ID, no password at all.
SSO / SAML
Sign in via your corporate login, e.g. Entra ID or Okta.
Wherever data must not leave the building
Law firms & consultancies
Today: Client data must not pass through a third party’s cloud AI.
With Quinta: Quinta stays entirely within your own network — client confidentiality is preserved.
Banks & insurers
Today: Regulation requires traceable, auditable AI usage.
With Quinta: Audit trail, RBAC and instance admin provide the governance required.
Industry & mid-market
Today: Trade secrets and design data should not flow to US clouds.
With Quinta: AI models run on your own hardware, from a single machine to a GPU server.
Public sector
Today: Data protection and sovereignty come before a feature race.
With Quinta: Self-hosted, an openly auditable core, full control over the location.
The questions CIOs ask first.
Every component — gateway, daemon, dashboard, model registry and database — runs on your hardware, in your network. There is no cloud portion, no telemetry to the outside and no provider that sees your requests.
From a single GPU workstation to AI appliances of the DGX Spark class (e.g. Asus GX10). Quinta detects NVIDIA, AMD and Intel GPUs automatically; every additional machine running the daemon becomes part of the platform on its own.
Through vLLM the entire HuggingFace catalogue — Llama, Mistral, Qwen, Gemma, DeepSeek and more. A lightweight model runner is available for a simple start. The lifecycle (download, start, health check, warmup) runs automatically.
Hardware and electricity. There are no token costs and no cloud bill — usage is unlimited and the budget is predictable.
Rollout onto existing hardware takes hours, not months. It is preceded by a 30-minute planning conversation about use cases and your compliance framework.
No. Quinta speaks the OpenAI API (/v1/chat/completions, embeddings, rerank, audio transcription, Responses API). Existing apps change base_url and API key — that is the entire switch.
Yes. Quinta runs models on your hardware and replaces cloud APIs where data sovereignty matters. Non-critical workloads can keep running in the cloud, independently — the decision is yours.
Behind Quinta stands twenty5ai — and Quinta is our own development. The entire sovereign platform is built in-house: gateway, daemon, registry, dashboard, access control, orchestration and the complete audit trail. For pure model execution we deliberately build on a proven open-source engine (Apache 2.0) so the whole open model catalogue stays available — everything that makes Quinta Quinta is twenty5ai's own development.
Quinta — Portuguese for a country estate: the place where your data is at home.