zeroroot
Sign inDesign partner
Solutions

By industry

Defense & national securityair-gapped, no egressFederal & public sectorinside your authorization boundaryFinancial servicesevery action attributableHealthcaredata stays in your boundaryCritical infrastructureoffline, segmented networksConsultancies & MSSPsper-client isolation, white label available

By workload

CI/CD & release managementreviewable commits, never the deployVulnerability & CVE responsealways-onCoding agents, under controlgrants, not vibesSecurity testingoffense and defenseIncident responsereplayableCompliance evidenceexportable
PlatformGibson RuntimeThe server your agents check in to. It holds each agent's identity and grant, runs missions, decides where work executes, and writes every event to the timeline.Gibson ConsoleThe web UI on top of the runtime. Create and watch missions, set grants and budgets, browse the knowledge graph, read traces, and replay any run.Execution environmentWhere untrusted work runs. Setec puts each tool run in its own microVM, on its own kernel, with the egress the component declared. The runtime dispatches into it and reads the result back.

How it works

Who can it act as? GrantsOnly what a named person delegated.What is it working on? MissionsA typed work graph, checked at submit.Where does the code run? Dispatch policyMicroVM, in process, or refused.What did it find? Knowledge graphHosts, paths and findings, per tenant.What did it do? ReplayAny run, rebuilt move by move.

Controls

What stops it? The six controlsEnforced, conditional, or structural, and where each one ends.
Open sourceADK Apache-2.0Build agents, tools and plugins. The gibson CLI.Zerocool Elastic License 2.0A coding agent built on opencode. A developer can drive it, but it is built to run on its own: the runtime dispatches work to it, from a mission or from a non-coding agent, and its changes arrive as commits a human reviews.Bridge Apache-2.0Brings any MCP-compliant tool in as a component.
PricingDocsSign inDesign partner

by industry

Regulated industries, inside your own boundary.

Each of these is led by the constraint that decides it. No customer, certification or deployment is claimed on this page.

Defense & national security

The workload cannot reach the internet, and the model cannot either. Both have to run inside the boundary you already hold.

  • One chart installs the runtime in your cluster, up to fully air-gapped
  • Local models: bring Ollama or a self-hosted endpoint, and no call leaves the boundary
  • The chart pins every image by digest at package time
  • An append-only timeline per tenant, exportable for review

Federal & public sector

The runtime installs inside your authorization boundary. It does not ask you to extend that boundary around someone else's cloud.

  • One Helm install into a cluster you already run
  • Your identity provider. Agents act on 55-second credentials
  • Every action traces to a named human's grant
  • Replay rebuilds a mission for a reviewer, step by step

Financial services

An agent that touches a regulated system has to be explainable afterwards, by someone who was not there.

  • An append-only timeline. Replay returns the same result every time
  • An agent can never hold more than the human who granted it
  • Deny wins wherever two policies disagree
  • A separate graph database per tenant, not a filter on a shared one

Healthcare

The question is not what the model was trained on. The question is whether the data ever leaves your boundary.

  • Self-hosted install. Your cluster, your keys, your region
  • Bring your own model, including one that runs locally
  • Declared egress, enforced at the microVM boundary
  • Per-tenant isolation at the database, not the query

Critical infrastructure

Segmented networks, long-lived equipment, and no tolerance for an agent that improvises.

  • Runs without egress. No phone-home in the install path
  • Untrusted work runs in a microVM, or the runtime refuses it
  • A mission has a type at submit, so a malformed mission never starts
  • Grants are explicit, reviewable and revocable

Consultancies & MSSPs

One engagement's data must never be visible from another. Every client wants their own evidence.

  • A separate graph database per tenant. No cross-tenant query path exists
  • Per-client grants, revocable the day an engagement ends
  • An exportable timeline per client
  • One chart installs the runtime in their cluster or yours. Or let zeroroot host it
  • White label available

Bring the workload you are not allowed to put an agent on.

Apply as a design partnerBack to the runtime