Who uses this, and for what.
From one developer's sandbox to governance across a fleet. Here is where teams usually start.
For developers
See what the agent is doing without slowing down. Run sw run -- your-agent with the open-source endpoint and keep working: the agent gets a sandbox with your repo mounted, and you see its file, process, and network activity live, with a session history you can replay. Free, no account, nothing leaves your machine.
Start with the Endpoint →For platform teams
Let everyone use agents, under one policy. Connect endpoints to a Server tenant: write policy once at the org level, tighten it per group or user, and push it everywhere. SSO on day one, telemetry in one console, and the dashboard flags endpoints that fall out of compliance. Team adds git provenance showing which agent session wrote what, with a 14-day trial.
See the Server →For teams running NVIDIA OpenShell
Keep OpenShell and attach SixWays to your existing gateway. SixWays compiles one policy into every sandbox, turns attempts to weaken a sandbox into approval requests, inspects model traffic on the middleware path, and records canary probe receipts of what the policy allowed and denied.
See SixWays for OpenShell →For agents off the laptop
Run agents in governed sandboxes inside your own cloud account, on a hardened Docker host, OpenShell, or OpenSandbox. You own the compute and the data plane; the same policy, approvals, and evidence apply there.
See Cloud Sandbox →For security teams
Kernel-level monitoring records system activity independently of the agent’s output. Policy is enforced at the sandbox and network layers, outside the agent process. Violations are blocked and recorded with full context for investigation.
How enforcement works →For teams shipping LLM features
Put the Firewall between your application and its model providers for inline prompt-injection, secret-leak and exfiltration scanning, with provider keys kept server-side. Use the Server’s campaign tooling and Arsenal’s test corpus to red-team your endpoints on a schedule.
See the Firewall →For regulated organizations
Keep custody of agent traffic, code, and telemetry: self-host the control plane and keep the Firewall in your own VPC. Licenses verify offline, retention is configurable, and fleet compliance reports map to SOC 2, HIPAA, NIST AI RMF, and ISO 42001. Public documentation shows which components run where.
Talk to us about Enterprise →For CI pipelines
Headless endpoints run CI agent jobs in the same sandboxes, under the same policy, as developer laptops. The audit trail records nightly and unattended jobs for review.
Talk to us about CI capacity →Get early access.
Tell us about your agents and your team, and we will set you up.