Zero Trust Dev

ZTD spins up disposable, isolated VMs to run untrusted, autonomous development work — AI coding agents in “YOLO” mode, sketchy builds, “just try this” — behind a zero-trust boundary. Spin one up, let an agent run wide open inside it, ship the result over git, throw the VM away.

The point: running an agent without approval prompts is only safe if a machine boundary, not a permission dialog, protects your host. ZTD makes that boundary cheap and disposable.

  • Zero trust — the guest is assumed compromised; blast radius ends at the VM.
  • Cattle, not pets — created per task, publishes over git, then destroyed.
  • One interface, many backends — local KVM (Linux) and vz (macOS/lima), Proxmox, EC2 (more to come).
  • Tooling in Docker — the host needs only Docker (plus a hypervisor for local).

Where to start

SectionFor when you want to…
Quickstartget a VM running in about a minute
Tutorialslearn ZTD by building something end to end
How-to guidesaccomplish a specific task
Referencelook up config keys, commands, backends
Explanationunderstand the zero-trust design
Examplesfollow a complete worked scenario

ZTD is pre-v1. Phases 0–2 are complete: all four backends (kvm, vz, proxmox, ec2) are validated end-to-end, and the ztd Go CLI now ships as a single installable binary with its runtime assets embedded — there is no source checkout to clone. See Install to get it, and the roadmap for what Phase 3 is adding.