Skip to main content
Each guide shows one working system, which workloads it is made of, why each one runs where it does, and how to run it. What you read is what you download.

Image studio

A Next.js and FastAPI app that generates images on a GPU, streams progress over a WebSocket and keeps a gallery.

Fine-tune an LLM with LoRA

Train an adapter on a GPU, score it in parallel against the base model, and serve it with vLLM.

Video to highlight clips

Transcribe talks from your S3 bucket, let an LLM pick the highlights and cut the clips in parallel.

Coding agents

Work with an agent on a devbox, turn GitHub issues into pull requests in sandboxes, and score the agent nightly.

Website change monitor

A scheduled job that checks pages in a real browser, has an LLM judge each change and alerts Slack.

Set up the examples

The download pins the SDK to the CLI you installed, and each guide’s commands run from the project directory. Commit the uv.lock that uv sync writes, since remote images install from it. GPU libraries install only in those images, so your machine needs no GPU. With several workspaces, uv run lazycloud workspace use <name> picks one. GPU examples need credit and GPU access, both under Settings, then Billing. Volumes and artifacts outlast a run, and each guide ends with the commands that remove them. The catalog also has quickstart, artifacts and all-workloads, each with its own README. lazycloud example download all writes every project.