Skip to content

Creating Agents ​

There are two real ways an Agent comes into existence today. There is no CLI command that creates an agent from scratch — machina agent only has list, get, run, and executions (confirmed directly in machina-cli's source; there's no agent create).

1. Create it in Studio ​

Studio → your project → Agents → New Agent. The form asks for exactly four things:

FieldNotes
TitleDisplay name.
NameSlug/identifier — set once, can't be changed after creation.
DescriptionFree text.
Statusactive or inactive.

That's the entire creation form. Two things it deliberately does not ask for at creation time:

  • Frequency (how often it runs) — set afterward, from the agent's detail page. See Configuring Agents.
  • Workflows (what it actually runs) — also attached afterward, not at creation.

So a brand-new agent exists as an identity + status shell; you wire it up to a schedule and to workflows as a second step.

2. Install a skill or template that provisions one ​

machina skills install <path> / machina template install <path> provisions cloud resources for your project — connectors, mappings, datasets, and (depending on the package) agents already wired to their workflows. This is the path for reusing a pre-built setup (e.g. one of the Agent Templates) instead of assembling an agent by hand.

bash
machina skills install <skill-path>
# or, for the older compatibility surface:
machina template install agent-templates/<template-name>

INFO

Whether a given template provisions an agent (versus just workflows/connectors) depends on that template's own packaged config. Run machina skills info <path> (or check the template's SKILL.md) to see what a specific package actually installs before you rely on it as your source of a scheduled agent.