<!-- source: https://docs.dreamlake.ai/lakeshore/examples @ 0.1.192 b2f973b1 -->
# Numbered DreamLake guides

Start with [01: register this computer and run hello](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/01-hello-ssh).
CLI 0.51.0 adds Linux `hosts enroll --local`; 0.52.0 adds macOS support, and
0.52.1 prints readable enrollment text by default (`--json` for programs). This computer can become an
execution host without SSH to itself. The guide verifies registration, a tracked
Python run and its persisted completion receipt.

The [numbered catalog](https://github.com/dreamlake-ai/lakeshore-examples) is the
single tutorial entry point. Earlier Lakeshore/Jaynes examples are merged into
these guides; retained source fixtures support regression tests, not a separate
onboarding path. Existing local queue examples retain their documented SDK limits.

| Guide | Included infrastructure |
| --- | --- |
| [17: EC2 CPU](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/17-ec2-cpu) | One explicitly configured EC2 machine, restricted network access and encrypted disk |
| [18: Development host](/examples/lakeshore/managed-agent-dev.md) | One EC2 host, user-selected Node/Codex/Claude pins and optional daemon startup |
| [20: GPU training](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/20-gpu-training) | One selected GPU instance; image/driver compatibility must be verified |
| [23: GCP machine](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/23-gcp-machine) | Private Compute Engine machine with OS Login and explicit host bootstrap |
| [25: Machine lifecycle](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/25-machine-lifecycle) | One machine with plan/apply, recovery and exact-state teardown |
| [27: Storage](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/27-storage-and-artifacts) | Private encrypted, versioned S3 bucket; no forced object deletion |
| [30: Batch/fleet](https://github.com/dreamlake-ai/lakeshore-examples/tree/main/30-batch-and-fleet) | Explicitly bounded fleet of one to four EC2 machines |

Copy a whole guide folder. Each carries its own Terraform, input example, helper,
provider lockfile and mock-plan tests, with no required sibling checkout. `check`
validates schemas and mocked plans without cloud mutations. `plan` saves a private
plan; explicit `apply` creates billable resources. No paid cloud apply was performed
for the recipe validation. Complete cloud workload acceptance remains separately
marked pending in each guide; successful Terraform validation is not execution proof.

## Dynamic Terraform data in Vault

Each infrastructure guide's `infra.py` requires a selected Vault prefix. It saves
and reads back the dynamic inputs, Terraform state and outputs under hashed entry
names, and writes a local receipt containing references only. Cloud credentials
use the provider's identity chain; personal tokens/private keys do not belong in
Terraform or cloud-init. A real local Terraform lifecycle verified Vault save,
readback, input recovery and teardown without provisioning cloud resources.

The helper backs up before apply and after success or partial failure. An
interrupted apply leaves its consumed plan for reconciliation; retry `backup`,
not the cloud mutation. Input restoration refuses an existing destination.
State restoration uses the canonical [Terraform Vault commands](https://cli.dreamlake.ai/vault)
with lineage/serial/hash checks and never pushes state automatically.

Vault is an encrypted snapshot destination, not an active Terraform backend or
distributed lock. These bounded tutorials use local authoritative state and a
single-computer lock. Larger/team deployments need an independently configured
locking backend. The existing 64 KiB serialized Vault entry bound still applies;
a failed backup leaves the state local and does not masquerade as success.
