AI
Sovereign AI Agents: What They Are and How to Run Them in Your Own Cloud
Sovereign AI agents are AI agents that work on your company’s data and systems while you stay in control of four things: where the data lives, which models read it, who sees each answer, and what the agent may do without a person saying yes.
That last point is what makes agents different from chatbots. A chatbot only talks. An agent reads your contracts, updates a ticket, drafts an e-mail to a customer or files a claim. Once software can act inside your company, “where is the server” is no longer the only question a regulator, a board or a security team will ask.
This guide is for the people who have to answer those questions: IT and operations leaders, compliance teams, and anyone choosing an agent platform in a regulated industry or a country with data residency rules.
Why companies are asking for sovereign AI now
Three things changed in the last two years.
- Agents moved from answering to acting. An assistant that drafts text is a productivity tool. An agent with access to your CRM, your mailbox and your finance system is part of your control environment, and auditors treat it that way.
- Data protection laws caught up. Saudi Arabia’s Personal Data Protection Law, the UAE’s federal data protection law and similar rules elsewhere put conditions on moving personal data across borders. Banking, insurance, healthcare and government regulators add their own rules on outsourcing and cloud.
- Most AI tools were built for one deployment. The popular assistants run in the vendor’s cloud, usually in the United States, with the vendor’s models. That is fine for many teams and impossible for others.
What “sovereign” has to cover
A vendor that says “your data stays in region” has usually answered one question out of six. Check all of them:
| What | The sovereign answer |
|---|---|
| Documents and files | Stored where you choose: your own bucket, in your region |
| The search index | In a database you choose, including your own |
| Models | Your choice per function: answers, reading scanned pages, search, extraction |
| Conversations and logs | Kept where the platform runs, for as long as you set |
| Who sees what | Each person gets answers only from documents they can already open |
| What agents do | Reading runs alone; writing, sending and spending wait for a person |
If any row is “the vendor decides”, the platform is not sovereign for that row.
The four ways to deploy AI agents
| Option | Where it runs | Good for | Watch out for |
|---|---|---|---|
| Public AI assistant | The vendor’s cloud and models | Individuals, public information | No control of location, models or retention |
| Managed platform | The vendor’s cloud, often one region | Fast start, most companies | Check the region, and what you can bring yourself |
| Your own cloud account | Your AWS, Azure or Google Cloud account, any region | Data residency, regulated teams | You run the infrastructure (or pay someone to) |
| On premises | Your own machines, models you host | Strictest rules, air-gapped networks | Highest cost, you own the hardware and upkeep |
Most companies do not need one answer for everything. A common pattern is the managed platform for a first pilot on non-sensitive knowledge, then the same product in your own cloud account once the pilot proves value and the sensitive documents come in.
The key word is same. If moving from the vendor’s cloud to yours means a different product, a rebuild and a new contract, you are choosing two vendors, not one.
Questions to ask every vendor
Take these into the first call. Good vendors answer them in one sentence each.
- Where is each type of data stored: documents, search index, conversations, logs and files?
- Can we bring our own model for answers, for reading scanned pages and for search, separately?
- Can the search index live in our own database?
- Does each employee get answers only from documents they can already open in the source system?
- Which actions can an agent take alone, and who approves the rest?
- Is every run, approval and settings change in an audit trail we can export?
- Does single sign-on, including SAML and SCIM, come with the product or cost extra?
- Is the product we pilot the same product that runs in our cloud or on our premises?
- What happens to our data when we leave?
What it costs
Sovereignty has a price, and it is mostly people and infrastructure, not licences.
- Managed platform: you pay for usage. This is the cheapest way to start.
- Your own cloud: you pay your cloud provider for the servers and database, the model provider (or your own hosted model) for model use, and a licence for the platform.
- On premises: add hardware, hosting staff and, for open-weight models, GPUs.
A useful rule: put only what must be sovereign in the expensive option. Public product information does not need an air-gapped server.
Where promptev fits
promptev is a platform for sovereign AI agents for business. Agents answer from your company’s data, see only what each person may see, and ask before they act.
- Three places to run, one product. promptev’s cloud, your own cloud account in any region, or your own premises. The product is the same in all three.
- Bring what you need. Your own models for answers, your own OCR and embedding models, your own database for the knowledge and your own storage bucket for files. Each is a separate choice, and what you do not bring runs on PromptevAI, promptev’s own models.
- Permissions in every answer. Each person gets answers only from documents they are allowed to open, in the source system.
- Approvals before actions. Per tool: run alone, ask once, always ask, or ask when a condition holds, such as an amount over a limit.
- Audit and single sign-on included. Every run, approval and change is recorded. SAML, OpenID Connect and SCIM come with every workspace.
You can start free with 1,000 credits and no card, or tell us what you need to keep where and we will tell you plainly which deployment fits.
Related reading
- Private AI: what it means and how to get it
- AI data leakage: the six paths and how to close them
- AI agent governance
- promptev vs Glean
Frequently asked questions
What is a sovereign AI agent?
An AI agent that works on your company's data and systems while you keep control of where the data is stored, which models read it, who can see each answer and what the agent may do without a person's approval. Sovereignty is about control, not only about location.
Is a model hosted in my country enough for sovereign AI?
No. The model is one part. Your documents, the search index, conversation history, logs and the keys to your other systems must also stay where you decide, and the agent's actions must be limited by your rules. A local model behind a foreign platform that stores everything else abroad is not sovereign.
Can sovereign AI agents run on premises?
Yes. The strictest option is to run the whole platform on your own machines, with models you host, so nothing leaves your network. It costs more to run than a managed service, so most companies start in their own cloud account in the region they need and move parts on premises only where a regulator requires it.
Does sovereign AI mean we cannot use models like GPT or Claude?
No. Sovereignty means you decide. Many companies use a commercial model for general questions and an open-weight model in their own environment for the most sensitive work. What matters is that the choice is per function and that it is yours, not the vendor's.
How do I check if an AI agent platform is really sovereign?
Ask where every type of data is stored (documents, index, chats, logs, files), whether you can bring your own models and database, whether answers respect each person's existing permissions, which actions need approval, and whether the same product runs in your cloud or on premises. Vague answers to any of these are a warning sign.
Faisal Saeed is Founder & CEO of promptev, the platform for sovereign AI agents for business: agents that answer from company data, see only what each person may see, and ask before they act.