Sovereign AI

On-premises sovereign AI, run inside your walls.

AI that runs in your data centre, under your jurisdiction, on your domain. Air-gapped when it has to be. For organisations in India and around the world that cannot send their work elsewhere.

Talk to us See the AI Factory

01 — What it means

Five questions decide whether AI is sovereign.

Sovereign AI has no single fixed definition. For a buyer it comes down to five questions, and whether you can answer each with “us”.

  • 1. DataWhere do prompts, documents and outputs live, and who can read them?
  • 2. ModelsDo you hold the weights, or rent access to a model you cannot inspect?
  • 3. InfrastructureWho controls the GPUs, the network and the building?
  • 4. OperationsWho runs it, from where and with what access?
  • 5. JurisdictionWhose courts and laws apply?

02 — Why on premises

For work that cannot go anywhere else.

  • Regulated and confidential dataCustomer records, legal files, source code and designs often may not leave the building.
  • AuditAn auditor can inspect a system you control.
  • ContinuityIt keeps working when a link to the outside is cut.
  • Control of changeA human approves what changes, and the audit trail is yours.

Many workloads are better served from a hosted endpoint. On-premises is for the ones that cannot be.

03 — What runs inside

The whole path, from GPU to API, on your site.

  1. Your toolsAny OpenAI- or Anthropic-compatible client, agent, IDE or script.
  2. GatewayKeys, sessions, admission and queueing, on your domain.
  3. ServingOpen-weight models on open-source serving engines.
  4. GPU fleetIn your data centre, on your hardware or ours.

Around them, an operations layer: observability, change control, scans and an audit trail. Open weights can be loaded and inspected inside the site, with no call to a third party.

04 — Air-gapped

Air-gapped first, not air-gapped later.

Air-gapped AI has no connection to the internet. It changes the design, not just the firewall: software and models arrive through controlled routes, observability runs inside the site, and every change waits for a named person.

Every MIRASTACK product has to work with no internet connection before it is allowed to use one. EGAProtocol, which we authored, puts identity, approval and an immutable audit trail into the protocol itself.

Day 2 operations

05 — How we deliver

Designed, built and run by one team.

  1. Day 0

    Design

    Workload sizing, hardware (yours or ours), power, cooling and network, model selection, security and compliance requirements.

  2. Day 1

    Build and go live

    Installation, serving stack, gateway and keys on your domain, acceptance testing.

  3. Day 2

    Run it

    Operated with the MIRASTACK operations stack.

See the AI Factory

06 — In India

In-country AI, with a team in Hyderabad.

MIRASTACK LABS PRIVATE LIMITED is headquartered in Hyderabad, Telangana. Many Indian buyers are regulated: financial institutions, public-sector bodies, telecom and critical-infrastructure operators. For them we build, run and keep AI factories in-country, on your hardware or ours.

The Digital Personal Data Protection Act, 2023 governs digital personal data in India, and some sectors have their own rules on data location and audit. Which obligations apply is for you and your counsel to confirm. Running inference in your own data centre in India makes “where does the data go?” easy to answer.

Agreements are with an Indian company and invoiced with GST.

07 — Worldwide

The same five answers, in any jurisdiction.

Each country has its own data protection law, sector regulators and view of cross-border transfer. The design does not change: compute inside the jurisdiction, keys and audit trail with the owner, operations a regulator can inspect.

Read the full guide

08 — Built for

Where cloud APIs are not an option.

  • Regulated financial institutions.
  • Defence and public sector.
  • Critical infrastructure operators.
  • Sovereign data centre operators.

09 — Commercials

Agreed per contract.

A sovereign AI system is scoped to the site and the workload. Hardware arrangement, models, term and what we operate for you are set per agreement.

10 — Questions

Sovereign AI, answered.

What is sovereign AI?

Sovereign AI is AI whose data, models, infrastructure, operations and legal jurisdiction are under your control or under a party you chose and can hold to an agreement. In practice it means you can say where every request goes, who can read it, who runs the system and which country's laws apply.

What is on-prem sovereign AI?

On-premises sovereign AI runs inside your own data centre, with GPUs, gateway, models and operations on your site and your domain. Nothing leaves your walls. It suits regulated or confidential workloads. It can be built on your hardware or ours, and be designed, built and run by MIRASTACK LABS as an AI factory.

What is air-gapped AI, and how is it different from private cloud AI?

Air-gapped AI runs with no connection to the internet or any outside network. A private cloud is still reachable from outside and is usually run by a provider. In an air-gapped site, models and updates arrive through controlled routes, observability runs inside, and all changes need human approval.

Does the DPDP Act 2023 require AI to run in India?

The Digital Personal Data Protection Act, 2023 governs digital personal data in India and lets the government restrict transfers to certain countries. It does not, on its own, say every AI system must be on premises. Sector rules and contracts may add requirements. This is general information; confirm your obligations with counsel.

Can MIRASTACK LABS deploy on-prem AI in India?

Yes. MIRASTACK LABS is headquartered in Hyderabad and builds AI factories that are run and kept in-country, on your hardware or ours. Agreements are with an Indian company and invoiced with GST.

What hardware does it need?

It is designed on Day 0 around your workload, site, power, cooling and network. You can bring your own GPUs or we can supply them, and in both cases the system sits in your data centre. The arrangement is set per agreement, along with models, term and the parts we operate.

Who runs it after go-live?

The team that built it, using the MIRASTACK operations stack: hardware-to-business observability, explainable root cause analysis, human-approved remediation, release risk checks, and supply-chain and cryptography scans. You can take parts in-house. Every action is identified, approved by a named person and recorded.

Which models can run on premises?

Open-weight models, selected with you and served on open-source serving engines inside your site. Because the weights are open, they can be loaded and inspected locally with no call to a third party, and the choice can change over time. Frontier proprietary models stay with their labs.

Talk to us

Tell us what you need to run.

Every agreement is shaped to the workload: seats, sessions, capacity, term and location.

Talk to us hello@mirastacklabs.ai