Self-hosted, so security says yes faster.
oneinfra was built for teams who can't send production telemetry to someone else's cloud. Everything runs in your environment, under your controls.
What actually runs in your cluster
No external control plane. No vendor data plane. Just two pods + your existing telemetry sources + the LLM provider you already use.
Footprint per cluster
| Component | CPU request | Memory | Storage | Replicas |
|---|---|---|---|---|
oneinfra-agent | 100m | 128Mi | — | 1 (HA: 3) |
oneinfra-control-plane | 500m | 512Mi | 1Gi PV | 1 (HA: 3 + Postgres) |
| Total (single-replica) | 600m | 640Mi | 1Gi | — |
Telemetry the agent reads (Prometheus, logs, K8s API) is whatever you already run — no additional storage required from oneinfra.
What your security team will ask about
Runs in your VPC
The entire platform — agent, control plane, and AI — deploys inside your own network. There is no oneinfra cloud in the data path.
No data egress
Telemetry, logs, and investigation results stay within your perimeter. Enterprise supports fully air-gapped operation with a self-hosted LLM.
Read-only by default
The agent observes with scoped, auditable permissions. No application data is extracted; remediations are explicit and guarded.
Encryption everywhere
TLS 1.3 in transit and AES-256 at rest, with your own keys in Enterprise deployments.
SSO, SAML & RBAC
Enterprise identity, role-based access control, and full audit logging of every action and AI investigation.
Bring your own model
Use OpenAI, Anthropic, AWS Bedrock, or a local model. Your prompts and context are never used to train third-party models.
SaaS AI-SRE tools ask you to trust their cloud. oneinfra doesn't ask.
Because nothing leaves your network, oneinfra clears the reviews that stall hosted competitors — data residency, vendor risk, and egress are simply not on the table.
-
✓No third-party data processorYour telemetry is never sent to a vendor for processing.
-
✓Your keys, your modelRun a local LLM for fully offline investigations.
-
✓Full audit trailEvery query and AI action is logged and exportable.
Bring it to your security team.
We'll walk through the architecture, deployment model, and controls in detail.