
Argent in your private cloud
Run Argent Strategic Intelligence on infrastructure only your organization uses, in your own data center or a dedicated cloud. The product and its whole data tier run in your cluster, and AI requests go only where you route them.
What private cloud means here
NIST defines a private cloud as infrastructure "provisioned for exclusive use by a single organization" that "may exist on or off premises." Source: NIST SP 800-145.
For Argent, that means a Kubernetes cluster or a Linux server your organization runs, in your own data center or a dedicated cloud, with a network that reaches the outside only where you allow it.
- Compared with [air-gapped on-premises](/platform/deployment/air-gapped-on-prem/): the same install, but the release can come straight from a registry instead of removable media, and agents can reach outside AI providers you approve.
- Compared with [public cloud VPC](/platform/deployment/public-cloud-vpc/): everything, including the database and the evidence store, runs in your cluster instead of a cloud provider's managed services.
Your hardware, your network, your keys. Argent checks and records every agent request that works through it, inside your walls.
What runs in your cluster
The Helm chart's self-hosted profile runs every part of Argent, including the data tier, inside your cluster.
| Part | What it is |
|---|---|
| The gateway | Every agent that works through Argent sends its requests here to be checked and recorded. It knows which agent sent each request and which person owns that agent. |
| The console | Where your team sets policy, watches requests and reads the evidence. It loads only from your install. |
| Database and cache | PostgreSQL and Valkey, running in the cluster. |
| Secrets store | Argent's built-in vault, OpenBao, running in the cluster. Or connect the vault you already run. |
| Evidence archive | Signed daily records on an object-locked store in the cluster, locked for seven years by default (your operators set the period). Your own S3-compatible store, such as MinIO, also works. |
| Sensitive-data detector | Ships in the signed release with everything it needs built into its image, so it makes no outside lookup. |
| Monitoring | Prometheus, Alertmanager, Grafana, Loki and Tempo ship in the release and run in your cluster. |
| Offline verifier | Checks the signed records on its own, inside the install. |
How it installs
One signed release, pulled from a registry and checked before it runs.
- Pick the form. The Helm chart for Kubernetes 1.34 or newer, with its self-hosted profile for clusters such as k3s and RKE2. Or the single-server bundle for one Linux host.
- Pick the registry. Pull from the Nyedis registry, where the single-server host check verifies image signatures, or mirror the release into your own registry. One setting points the install at it:
global.imageRegistryfor the chart,ARGENT_REGISTRYfor the single server. - Size it. Four sizing profiles, from small to extra large, set resources, storage and the default request-rate limits.
- Start it. The database migrates itself on start. On the single server, secrets are generated on the box for each install and never printed.
- Update it. The next signed release goes through the same checks. On the single server, the host check rebuilds the setup from the new release instead of keeping the old one.
Measured: on one cloud server, the single-server bundle was serving 100 seconds after unpacking, with images already in a private registry.
Connects to what you already run
Argent plugs into the systems your private cloud already has. Each connection exists only when an admin adds it.
- Your vault. Provider keys stay in HashiCorp Vault or OpenBao, CyberArk Conjur, Infisical or AWS Secrets Manager. The gateway adds the key to each call on its way out, so agents never hold it. See keyless agents.
- Your directory. People and groups come from Okta, Microsoft Entra, Keycloak, Auth0 or LDAP and Active Directory, so every agent is tied to a real owner. See identity tethering.
- Your AI models. Routes point at local engines such as Ollama, vLLM and LM Studio, or at outside providers you approve, such as OpenAI, Anthropic, Azure OpenAI, Amazon Bedrock and Google Gemini. Only approved models can answer. See model control.
- Outbound control. Connector traffic goes through a guarded client, and an allowlist can make it deny-by-default.
More in our integrations.
Built to stay up, and to fail safe
The chart ships the defaults a production cluster needs.
- Can grow with demand. With KEDA in your cluster and autoscaling turned on, Argent starts extra copies of its gateway when more agents send requests, and they go away when things quiet down.
- Spread out. Disruption budgets and topology spread keep copies apart and limit how many go down at once during maintenance.
- Two doors. Only the gateway and the console face the network. The chart's network policies keep everything else inside the cluster, where the cluster's network plugin enforces them.
- No request without its record. By default, if the audit buffer fills, the gateway refuses new requests instead of serving them unrecorded. Your operators can choose otherwise, and then the gap is marked on the signed record.
- Clean restarts. On shutdown the gateway stops taking new work, lets work in flight finish and flushes its records.
Where it fits, and how a team starts
Private cloud suits a team that runs its own infrastructure and wants Argent's data to stay there. Compare it with air-gapped on-premises and public cloud VPC on the deployment options page.
- Goals: sovereign AI, data loss prevention, agent identity and ownership, audit and accountability and all goals.
- Use cases: securing AI coding agents, rolling out copilots and chat safely, controlling agent tool access and AI in regulated and air-gapped environments.
- Industries: Financial Services, Healthcare, Insurance, Public Sector and all industries.
- Install. Put the chart or the single-server bundle on your infrastructure, from the Nyedis registry or your mirror.
- Connect. Add your vault and your directory, then point your agents at the gateway and your routes at the models you approve.
- Watch, then enforce. Start in watch mode, which records what Argent would have done while nothing is stopped, then turn on enforcement. Take the walkthrough or browse all capabilities.
Want this running on your own use cases?
All downloads