How it worksPlatformUse casesSecurityManifestoDocsStart building
Security & trust

Your code and data,
contained.

Every build runs in an isolated sandbox. Nothing you bring is used to train public models. Verification isn't just about quality — it's about never shipping something you can't trust.

Sandboxed builds

Each build runs in its own isolated microVM. Code executes there — never on shared infrastructure — and the environment is destroyed when the build ends.

Encrypted end to end

Data is encrypted in transit and at rest. Secrets and environment variables are injected at deploy time and never written to logs.

Never trains public models

Your prompts, code, and data are yours. They are never used to train shared or public models, full stop.

Full auditability

Every agent action and tool call is traced. Each build ships with a complete event log — who did what, when, and why.

Your code, your repo

The full source is yours. Push it to your own GitHub, deploy to your own accounts, and walk away with everything.

Human in the loop

Gate any deploy or critical action behind your approval. Agents propose; you stay in control of what ships.

Containment by design

The build can't reach past its sandbox.

Agents write and run code inside a disposable environment with a strict path and network boundary. Protected paths reject writes, reads outside the workspace are blocked, and nothing persists beyond the build unless you ship it.

  • Workspace-relative paths only — no escaping the sandbox
  • Whitelisted commands; no shell expansion, no privilege escalation
  • Disposable microVMs, torn down after every run
sandbox · build #4827
// isolated microVM
fs.write ✓ workspace-only
net.out ✗ blocked
secrets ✓ inject-at-deploy
teardown ✓ on exit
Compliance

Built to meet you where you operate.

We're building toward the standards enterprise teams expect. Talk to us about your requirements.

SOC 2 — in progress Encryption in transit & at rest Data isolation per build SSO — on the roadmap Audit logging