RelayOne sits in front of every AI call your company makes. It tags where the request came from, where it is going, and what it is allowed to carry. It can rewrite, reroute, or block in flight without breaking the workflow that triggered it.
For compliance leads · platform engineers · heads of AI
Use one governance layer to decide who can use which model, which requests stay in-region, which prompts are rewritten inline, and which calls must be rerouted to an internal agent instead. The proof trail lives in the application and database layer, so the guarantees hold in cloud, on-prem, sovereign, and air-gapped deployments.
Right now, somewhere in your company, a support rep is pasting a customer transcript into a chatbot. An engineer is handing a database schema to a public model. A new hire just signed up for their fourth AI tool with a company card.
None of them are trying to cause a problem. They’re trying to get their work done. The real issue is there’s no sensible middle ground between “block everything” and “hope for the best.” RelayOne is that middle ground.
Composite based on customer interviews — not attributed quotes.
This chain is the structural proof behind RelayOne. A request can pass through nine nodes and you can prove afterward exactly what each node did. Break any link and the rest stops verifying.
You don’t need to understand the plumbing to understand the change. Four ordinary moments, on either side of switching RelayOne on.
Most governance products have an on-prem mode that's missing half the features. RelayOne's architectural bet means the feature set is identical across postures. Compare your posture to a cloud-dependent stack.
Every LLM request is a packet crossing a network. Most organizations can't name which ones leave their boundary, which carry private data, and which could be served by internal infrastructure. RelayOne maps the flow, classifies by destination, and — on demand — locks egress and reroutes through RelayGate with PII redaction.
Deploy policy updates atomically across tenants. Rotate credentials through the vault. Watch deployment status across regions. RelayOne's control plane treats every RelayGate instance as part of a coordinated fleet.
A request leaves your network headed for a public provider. RelayOne identifies the agent, inspects the payload shape, strips the PII, and redirects the call to an internal R1 agent running inside RelayGate. The employee's workflow keeps running. The sensitive data never leaves. This is what "safe AI use" means operationally.
Every employee session, every agent, every internal tool call is an object in RelayOne. Per-user policy sets. Per-team budgets. Per-category allow/reroute/block. Network telemetry, identity, and commerce on a single timeline you can actually act on.
Toggle the components you need. RelayOne assembles the evidence bundle: audit chain snapshot, policy decision log, fleet configuration digest, deployment provenance, signed receipt archive. Download a PDF. Or pass a signed bundle to your auditor.
Not features. Specific architectural decisions that make RelayOne deployable where most governance products cannot go.
Every other AI governance product relies on the cloud provider for the audit backstop: CloudTrail for the event log, KMS for key custody, S3 for retention, cloud-provider IAM for authorization. RelayOne carries its own chain end-to-end in the application and database layer. Turn off the cloud. The guarantees hold. This is why air-gapped, sovereign, and regulated deployments are possible.
RelayOne governs on request shape, origin, destination, metadata, and policy context, not on prompt content. Your customer data stays in the request path. The governance layer evaluates flow properties without needing to see the payload. Sovereignty rules and compliance guarantees hold without adding a data-inspection surface.
Most organizations run two products: one for fleet deployment and credential rotation, one for compliance reporting and audit. They drift. RelayOne runs both surfaces on the same substrate. Deploying a policy updates both the enforcement plane and the evidence surface atomically.
Most audit tools produce a CSV export. RelayOne produces a structured bundle: the audit chain snapshot, the policy decision log, the fleet configuration digest, the deployment provenance, the signed receipt archive. Assembled on demand, rendered as a PDF or passed as a signed bundle. Your auditor reads a deliverable, not a dump.
The request lifecycle has clear phases: ingress, identity resolution, scope check, budget check, sovereignty gate, dispatch, response, evidence emit. RelayOne evaluates policy at each phase. A sovereignty violation at the dispatch phase blocks the dispatch; an identity check failure at the start blocks everything. Policy is not a log reader. Policy is a gate function evaluated before action.
Hover any tile to trace the paths. Click a composition below to highlight exactly which products cooperate to solve a specific control problem.
Status (Q3 2026 roadmap): RelayOne ships as a standalone gateway today. The "RelayOne in front of R1", "RelayOne reroute through Veritize", and "TrueCom DID issuance at tenant provisioning" compositions shown below are on the integration roadmap; cross-product control-plane wiring lands progressively. Fleet governance (RelayOne → RelayGate hot policy reload) is partially wired today. Concepts & status →
RelayOne, RelayGate, Heroa, TrueCom, CloudSwarm, and CanadianClaw are commercial. R1 + STOKE is Apache 2.0. The substrate is open. The gateway is what ties it together.
Full pricing and feature matrix on /pricing. Quick preview below.
See full comparison on /pricing. On-prem and sovereign details on /on-prem.
Walkthrough if you want to see it run. Sovereign consultation if you already know on-prem is on the table.
We show the tamper-resistance chain, the fleet control plane, and the evidence bundle assembly on your actual use case.
For regulated, air-gapped, or sovereign deployments. We cover architecture, deployment, and contractual structure.