We build and operate the infrastructure that other teams use to run large language models, and we publish what we learn doing it.

Documentation status

This site is an early version

The structure is complete and navigable, but most content is still marked draft — meaning the page states its scope, learning objectives and planned outline rather than containing finished technical content.

Draft pages are labelled in the sidebar and on the page. Where a value or procedure specific to the llm-inference service has not been published, the page says so explicitly and uses a visible placeholder rather than guessing.

Contributing

The whole site is Markdown in one repository, built with Hugo. Contributions we particularly want:

  • Corrections. Technical errors, broken links, outdated statements.
  • Filling in draft pages. Each draft page carries a recommended outline to work against.
  • Service specifics. Confirmed endpoints, setting names and commands that currently sit behind placeholders.
  • Blueprints. New blueprints, or improvements to existing ones.

The repository README.md documents the local setup, the archetypes for new pages, and the conventions to follow.

Tip

When filling in a draft page, keep the front matter status field accurate — change it from draft once the page contains real content. The sidebar and search results both use it.

Conventions used on this site

  • Placeholders appear as <EXAMPLE_VALUE> for any value that depends on your deployment or has not been published.
  • Draft badges mark sections that are outlines rather than content.
  • Callouts carry notes, tips, warnings and important caveats.
  • Concepts are explained once, in the handbook, and linked from everywhere else.