Skip to content

Onboarding checklist

The order below is the order the wizard uses, and it is not arbitrary: each step makes the next one possible.

  1. Connect your agent — mint an API token and point your agent's MCP client at Hive.
  2. Install the GitHub App — before registering a project, not after. See 03-github-app-and-projects.
  3. Register the project — tell Hive which repository.
  4. Runner — confirm you have compute. Connect a machine of your own with the one-line install, or run on the booster allowance your tenant holds.
  5. First run — push a .hive/main.star and watch it go green.
  6. Credentials — model keys and pipeline secrets. Only needed for *agent* work; ordinary pipelines run without them.
  7. Team — invite the rest of your people.

Where you are right now

Ask Hive rather than guessing:

get_onboarding_state

It answers with has_project, has_cluster_online, has_run, has_green_run and names the next action. Call it whenever you are unsure what remains — it reads live state, so it is always more current than this document.

The two things that block people

The GitHub App must be installed on the account that owns the repository. Registering a project does not install anything and does not verify the install; the two are separate actions and the wizard puts the install first for that reason. A project registered without an installation looks fine until the first webhook never arrives.

Agent launches need your own model key. A tenant run resolves *your* tenant's model credential and never falls back to the operator's. With none configured, an agent launch fails naming the missing variable. Ordinary CI — build, test, lint — needs none of this and works immediately.

This page is the hive-docs knowledge collection your agent reads inside the product, published verbatim. Markdown version