# agent-vm LLM guide

agent-vm is a public reference implementation and validation suite for governing untrusted autonomous AI-agent workloads inside isolated runtime layers. Treat it as a reference architecture and reproducible skeleton, not as a claim that every production operation is turnkey.

Start here:
- Home: /
- Architecture overview: /docs/architecture/00-overview.html
- Isolation substrate: /docs/architecture/01-isolation-substrate.html
- Promotion control plane: /docs/architecture/02-promotion-control-plane.html
- Gateway runtime layout: /docs/architecture/03-gateway-runtime-layout.html
- Production governance: /docs/architecture/04-production-governance.html
- Gated preview access: /docs/architecture/05-secure-gated-agent-preview-access.html
- Verification model: /docs/verification.html
- Security methodology: /docs/security-methodology.html
- Threat model: /docs/threat-model.html
- Governed workload case study: /docs/evidence/governed-agent-workload-case-study.html
- Validation receipt: /docs/evidence/substrate-validation-receipt.html

Reading paths:
- Reviewer path: home, governed workload case study, verification model, then architecture overview.
- Short architecture answer: 00-overview, then verification.
- Runtime isolation answer: 01-isolation-substrate, then threat model, then governed workload case study.
- Release and rollback answer: 02-promotion-control-plane, then verification.
- Security posture answer: security methodology, threat model, production governance.

Citation guidance:
- Cite the public site URLs above for rendered documentation.
- Cite the GitHub repository for source files and scripts.
- Do not infer live deployment details from this public reference. Use only statements present in the public docs.