IAaaS Sovereign AI infrastructure

All your AI services for your project, under one subscription

For teams that need serious AI and cannot send their data to a public cloud. Your developers keep the tools they already know, and the data stays where you decided.

Book a demo See what it serves
  • 6 AI services inference, RAG, embeddings, agents, voice, OCR
  • 4 Deployment modes managed, sovereign cloud, on-premise, air-gapped
    Monaco Cloud Hosting or your cloud, your datacentre, or air-gapped
  • 2 Sovereignty levels Monaco jurisdiction, or the European Union
    Every call Audit immutable, and exportable as a file
  • Still works Your existing code OpenAI-compatible, so only the address changes

The service catalogue

Every generative AI service, on one sovereign platform

Six services, one API, one bill, one audit trail. It is the layer the Altores AI agents themselves call, opened so you can build on it directly.

What do you want to build?

Pick an outcome. The services it needs light up below.

LLM as a service Chat and reasoning models your developers reach the way they already do. Switching means changing one address, not rewriting anything. You skip a rewrite, a second SDK to learn, and the GPU provisioning, model serving and token accounting underneath
RAG as a service Retrieval-augmented generation over your own documents, classic or graph RAG, with citations in the answer. You skip a vector database, a chunking pipeline, a retrieval layer and citation tracking
Embeddings API Vector representations for search, classification and deduplication, served from the same platform. You skip a second vendor, a second contract and a second data boundary
Agent and assistant builder Assistants and autonomous agents built on the catalogue: tools, memory, guardrails and handover. You skip an orchestration framework, a tool registry and a state store
Voice-native AI agents Agents that answer on a phone line, with speech handled natively rather than bolted onto a text model. You skip speech-to-text, text-to-speech and turn handling stitched together
Document OCR PDF, PNG and JPG in, structured Markdown out, tables and headings included. You skip an OCR vendor and the parsing layer around it

How it works

One call in, a grounded answer out, and a line in the audit log

The same call, in two sectors where the answer has to be defensible. Watch which service takes it, what it reads, what it answers, and what it books. Nothing leaves the perimeter chosen for that deployment.

Book a demo

AI services

LLM inference
RAG
Embeddings
Agents
Voice agents
Document OCR

Watch one work

0Calls
0Usage
IdleStatus
  1. AI service
  2. Rules checked
  3. Action
  4. Answer

API & audit log

  • OpenAI-compatible request and response
  • Streamed back token by token
  • Usage booked against the calling key
  • Every call written to the audit log

Client apps

Clinical portal
Client records
Billing
Document intake
Your own app
Agent runtime

Running it

The API answers. The platform accounts for it.

Who is allowed to call what, how much it has cost, and a record that survives an audit. This is the part that separates a model you tried from a platform you run.

  • Usage metering and quotas

    Token consumption and OCR page counts tracked per client, per model and per billing period, with monthly quotas and either a hard block or pay-as-you-go overage.

    within quotaapproachingceiling reached

  • Immutable audit log

    Every API call, authentication event and admin action recorded with timestamp, client identity, model, token count and outcome. Exportable for a regulator or an internal review.

  • Access control

    Keys scoped per client and per model, so a project team can be given exactly the surface it needs and nothing beyond it.

Reported on

Every report exports to CSV, over any date range.

  • Consumption tokens and pages, by client, model and period
  • Quota usage against allowance, and what happens at the ceiling
  • Access which key called what, and when
  • Errors rejected calls, with the reason
  • Latency response times per model and per endpoint
  • Audit export the full log, as a file a regulator can read

More than access credentials

Engineers size it, deploy it, and connect it to what you already run

Nobody here hands you a key and calls it a deployment. The work is sizing the hardware for your load, installing the platform in the mode your rules require, onboarding the models you need, and wiring the API into your systems.

Talk to an engineer

  1. Requirements

    What your rules require, what you intend to build, and which of the four modes that lands you in.

  2. Sizing

    Hardware sized for your expected load, or the managed capacity reserved for it on Monaco Cloud.

  3. Deployment

    The platform installed in the chosen mode, air-gapped delivery included when there is no external network.

  4. Model onboarding

    The models you need served and tested, including any you bring yourself.

  5. Integration

    Keys, quotas and audit configured, and the API wired into the systems that will call it.

  6. Handover

    Your team trained on the console, the metering and the audit export, and running it themselves.

Questions

What technical and regulated buyers ask first

The eight that decide an evaluation, answered without the marketing layer.

A third-party AI API decides where your data goes. IAaaS is a private LLM platform that lets you decide: managed by Altores in Monaco under Monaco law, on your sovereign cloud, in your own data centre, or fully air-gapped. Nothing is ever forwarded to an external AI provider in any of them. Existing integrations keep working unmodified, and moving between those modes later does not change a line of your code.
Yes. IAaaS exposes an OpenAI-compatible REST API, so any application already using the OpenAI SDK switches by changing the base URL and the key. Streaming, multi-turn conversations and multimodal inputs are supported.
LLM inference for chat and reasoning, retrieval-augmented generation over your own documents in classic or graph form, an embeddings endpoint, a builder for AI agents and assistants, voice-native agents, and document OCR that returns structured Markdown. All of them behind the same keys, the same metering and the same audit log.
That is what it is for. IAaaS is the layer the Altores AI agents themselves call, exposed so you can call it the same way: authentication, quotas, model routing, usage accounting and audit logging are already there. A project team integrates and ships, without standing up inference servers, a billing pipeline or a compliance trail first.
In the managed mode the platform runs on Monaco Cloud and the data is held under Monaco jurisdiction, with GDPR-aligned processing terms; Monaco Cloud itself holds HDS certification, the French standard for hosting health data. In the self-hosted modes, sovereign cloud, on-premise and air-gapped, the data stays inside your own perimeter, Altores holds no copy, and data residency is whatever your own infrastructure already satisfies. Every mode writes the same immutable audit log.
The platform meters usage per client, per model and per billing period, with configurable monthly quotas and either a hard block or pay-as-you-go overage. Usage data is visible in the console and exports to CSV over any date range.
Yes. It deploys in environments with no external network access, with images delivered offline. All inference happens locally, and nothing in the platform requires an internet dependency to serve a request.
It runs on a single GPU server for a small deployment and scales to a cluster. In the managed mode there is no hardware for you to buy at all. Sizing is part of the deployment work rather than a number quoted before anyone has seen your load.

Next step

Build on a stack you are allowed to use

Tell us what your rules require and what you want to build. We size the deployment, onboard the models and connect the API, and your team starts against an endpoint that already exists.

Book a demo