Runs inside your perimeter

We don't ask you to send us anything. We deploy into your VPC.

Onepane is VPC-deployed root cause analysis, not SaaS. Your telemetry, logs, traces, configuration and change data never leave your account. There is no log shipping to a vendor cloud and no production data egressing to a third-party model API. The RCA is generated where the data lives, under your keys, inside your controls.

Why this matters most

"We are not sending our production logs to your cloud." Agreed.

The number-one blocker on every enterprise AI-operations deal is data residency: "we are not sending production logs to your cloud, and we are certainly not sending them to an LLM provider." At a bank, a health system, a utility or a government-adjacent buyer, that objection ends most conversations before anyone looks at the product.

Every SaaS tool in this space is disqualified in week one on that sentence. Onepane isn't, because the objection is our architecture. If your InfoSec team raises it unprompted, they have just told you which vendors are left.

Data flow

What leaves our environment? Answered precisely, in writing.

Stays inside your VPC
Telemetry
Metrics, logs, traces, queried in place from your observability tools. Never shipped.
Change and configuration data
CI/CD, IaC state, config, cloud-provider events, read where they live.
Topology and identity
Kubernetes, cloud control planes, DNS, IAM, tags, read-only.
Model inference
Runs inside your boundary. No production data is sent to a third-party model API.
The artifacts
Generated in your VPC, delivered into your ITSM. You own everything we produce.
Crosses the boundary, disclosed unprompted
Service heartbeats
Liveness of the deployment itself, so we can meet the SLA.
Aggregate service metadata
Counts and timings about our own service, never customer telemetry.
Support access, when you grant it
Scoped, time-boxed, logged, customer-revocable access for our expert pod.

If anything egresses at all, we say so before you ask. Being caught understating this once would cost us the deal and the reference, so we don't.

Deployment models

How do deployment models differ for incident-investigation AI?

SaaS, multi-tenant "Bring your own cloud", vendor-managed Customer VPC Air-gapped
Where your telemetry lives Vendor cloud (multi-tenant) Vendor-managed account in your cloud Your VPC, your account Your VPC, no internet path
Who holds the keys Vendor Shared / vendor-operated You You
Model inference location Vendor / third-party API Varies, often external Inside your boundary Inside your boundary
What egresses Everything the tool reads Depends on vendor Heartbeats + aggregate service metadata, disclosed Nothing (offline update path)
Typical security-review outcome at a bank Disqualified on data residency Long review; often conditional Approved with standard controls Approved for the most restricted estates
Onepane , , Default Available

Full vendor-by-vendor view: incident-investigation AI tools compared by deployment model →

Security and architecture review

This is our home field. Most vendors never reach this meeting.

We open with the deployment model, not the product. Here is what we bring, so your InfoSec team can defend us internally instead of blocking us.

01VPC reference architecture diagram (AWS, Azure, GCP)
02Data-flow diagram showing exactly what does and does not cross the boundary
03Model-hosting statement: which models run where, and how they are updated
04Key management and encryption at rest / in transit
05Expert-pod access model: scoped, time-boxed, logged, customer-revocable
06SOC 2 status and most recent third-party penetration-test summary, provided in security review
07Incident-response and breach-notification terms
08Source escrow and data-portability terms

The operating model, how our expert pod accesses the tenant, what leaves it, and who is on the hook when the SLA is at risk, is defined in the contract before a security reviewer asks.

FAQ

Data residency and deployment, the questions.

Does our data leave our environment?

No production data does. Onepane deploys into your VPC; logs, traces, change and configuration data are queried where they live, and model inference runs inside your boundary. What does cross the boundary, service heartbeats and aggregate metadata about our own service, is listed in writing before you sign, and never includes customer telemetry.

Can AI root cause analysis run inside our VPC without sending logs to a vendor?

Yes. That is Onepane's default and only deployment model. The agentic investigation, the model inference and the artifact generation all run inside your account. Nothing is shipped to a vendor cloud or a third-party model API.

Why do security teams block SaaS incident-investigation tools?

Because those tools require production logs, often containing customer data, credentials and internal topology, to be sent to the vendor's cloud and frequently to an external model provider. For a bank, health system or utility that is usually a disqualifier before the product is evaluated. A VPC deployment removes the objection instead of arguing with it.

How does a managed service work if it runs in our VPC?

The software runs in your account; our engineers operate it through an access path you grant, scoped to the deployment, time-boxed, fully logged, and revocable by you at any time. What our expert pod can see, and what leaves the tenant, is defined in the contract before a security reviewer asks.

Is air-gapped deployment supported?

Yes, for the most restricted estates: no internet path, offline update process, all inference local. Discuss it during the architecture review; it changes onboarding time, not capability.

You're a startup, what's the vendor risk?

Three mitigations. The deployment lives in your VPC, so there is no service to go dark and none of your data held elsewhere. Source escrow and data portability are available in contract. And we are US-headquartered with the team in California, so support and accountability are in your timezone and jurisdiction.

Start with the reference architecture.Then the replay.

Ask for the VPC reference architecture and data-flow diagram, then send us your last 90 days of Sev1s. Two weeks, no cost.