
ARGENT FORSOVEREIGN AI
Master Control System · by Nyedis
Your goal: keep your AI under your own control
If your goal is sovereign AI, the question is simple: where does your AI run, where do your prompts and records go, and who decides? Gartner expects most organizations in Europe to answer it in writing within a few years.
- Gartner predicts that by 2029 more than 90% of organizations in the European Union will have a digital sovereignty strategy, as they seek self-sufficiency across the whole stack, "including data, technology, operations and AI."¹
- Gartner forecasts worldwide sovereign cloud infrastructure spending of $80 billion in 2026, up 35.6% from 2025, with governments the main buyers, followed by regulated industries and critical infrastructure.²
AI agents send your data to a model with every request. Sovereign AI means the checkpoint, the data and the evidence stay where you put them. Argent was built to run that way.
Seven questions sovereignty asks, and where Argent answers
Sovereignty is a set of plain questions. Argent answers each one with a setting your team controls.
| Question | How Argent answers it |
|---|---|
| Where does the checkpoint run? | Inside your environment: air-gapped on premises, private cloud or your public cloud VPC. The gateway, console and sensitive-data detector all run in your install. |
| Where do prompts go? | Only where your routes point: a local model such as Ollama, LM Studio or vLLM, or an outside provider you approve. Once enforced, only approved models can answer; a request for any other model is refused. |
| What does an outside model see? | What your policy allows. Social Security, card and account numbers can be masked before a request leaves and, if you switch restore on, put back in the answer the person sees. |
| Where are the records kept? | On storage you pick: inside the install by default, or your own object-locked store in AWS, Azure or Google Cloud. |
| Whose keys? | Yours. Provider keys stay in your vault, and by default agents never hold them. |
| Who can check the evidence? | Anyone you hand it to, on their own machine, with no call to Nyedis. |
| Does it phone home? | No. Argent's own services open no outside connection by default. |
The path: from a sealed network to enforced rules
Each step is a choice your team makes, in order, and no AI request goes anywhere until your team sets a route for it.
- Choose where Argent runs. One Helm chart for Kubernetes covers Amazon EKS, Azure AKS, Google GKE and self-hosted clusters such as k3s and RKE2, and a single-server bundle covers one Linux host. Compare deployment options.
- Carry it in. For a sealed network the release arrives as one signed bundle on removable media. The import checks the release signature, every file and every image, stops on any mismatch, and loads your own registry. See Air-Gapped Install.
- Keep the data tier on your side. The self-hosted profile runs the database, cache, secrets store and object-locked evidence store inside the cluster. Or point it at your own managed services in your own cloud account.
- Route to the models you approve. Point routes at local engines inside the network, or at approved outside providers such as AWS Bedrock and Azure AI Foundry. Once enforced, model access is deny by default: a request for a model nobody approved is refused. See Model Control.
- Mask what must not leave. Sensitive Data Redaction masks or blocks sensitive data in the request before it reaches the model.
- Watch, then enforce. Start in watch mode: requests pass unchanged while Argent records what each control would have done. Then switch controls on one at a time.
What stays inside, and what leaves only by choice
An air gap is an interface where two systems "are not connected physically" and data moves "only manually, under human control."³ Argent was built to run there: everything it needs runs inside your own network.
| Part or connection | Default |
|---|---|
| Gateway and console | Inside the install. The console loads nothing from another site; fonts and the world map are bundled. |
| Sensitive-data detector | Ships in the signed release with its language model built into the image, so it needs no outside lookup. |
| AI models | Wherever your routes point. Local models are metered too, at a cost of zero. |
| Argent's own AI | Not needed. The few features that use a model are optional, and you pick the model, even a local one. |
| Model price feed | Off. It runs only if an operator turns it on, and new prices wait for an operator's approval. |
| Connectors (identity directory, alerts, vault) | None at install. Each exists only when an admin adds it, and an allowlist can make identity, vault and tool-server connector traffic deny by default. |
| Certificates and tracing | A local certificate, no public certificate authority. Traces go to the collector inside the install. |
Where your company lets a connection out, the same gateway reaches outside AI providers, and each request is checked on its way out.
Your keys, your vault, your records
Data residency is not only about prompts. It is also about who holds the keys and where the evidence lives.
- Provider keys stay in your vault. Argent reads them from your own secrets store, such as HashiCorp Vault, OpenBao, AWS Secrets Manager, CyberArk Conjur or Infisical, and adds the key at the gateway. By default, the agents that work through Argent never hold a provider key. See Keyless Agents.
- The evidence archive sits where you put it. On a self-hosted install it is the object-locked store inside the cluster. In a cloud, it can be S3 Object Lock, Azure Blob storage with immutability or Google Cloud Storage, in your own account. On Azure and Google Cloud the archive signs in with the workload's own identity, with no stored key.
- Every record names the provider. Each signed record carries the provider the request went to and the region label of the record chain it was written into.
- Every request has an owner. Each registered agent is tied to a named person. See Identity Tethering.
Proof that stays on your side
Sovereignty has to be shown, not just stated. Argent's evidence is checked without trusting Nyedis.
- Signed, tamper-evident records of every request and decision, allowed or blocked, with the agent, its owner and the reason.
- An offline verifier that runs inside your install, re-checks the sealed evidence every hour and keeps its own log of each check.
- An evidence package an examiner checks on their own machine. See One-Click Audit Evidence.
- Replay of what an agent asked and what came back. See Session Replay.
What the rules already ask for:
- For organizations subject to the EU General Data Protection Regulation (GDPR), Article 44: a transfer of personal data to a country outside the European Economic Area, or to an international organisation, "shall take place only if" the conditions of its transfer chapter are met.⁴ Argent's routes and masking decide what reaches each provider, and each signed record names the provider the request went to.
- For US federal information systems, NIST SP 800-53 control SC-7(5): deny network traffic by default and allow it by exception.⁵ Argent's own services open no outside connection by default.
For the full audit picture, see the compliance evidence and audit and accountability goals.
Where to go next
Sovereign AI is the ground the other goals stand on. These pages take it further.
- Capabilities on this path: Air-Gapped Install, Model Control, Sensitive Data Redaction, Keyless Agents, One-Click Audit Evidence, Identity Tethering and Watch Before You Enforce. All capabilities.
- Related goals: see all AI in your company, data loss prevention, model control and routing, compliance evidence, audit and accountability, agent identity and ownership, threat defense and budget and token spend. All goals.
- Use cases: AI in regulated and air-gapped environments, Securing AI coding agents, Rolling out copilots and chat safely, Governing customer-facing AI assistants and Controlling agent tool access.
- Industries: Public Sector, Financial Services, Healthcare, Insurance, Retail and all industries.
- Where it runs: air-gapped on-premises, private cloud or your public cloud VPC. Compare deployment options.
- Connects to: 17 AI providers and 65 agent tools and frameworks. Browse our integrations.
- See it yourself: take the walkthrough or download the data sheets.