Skip to main content
An Agent is the scheduling primitive in Machina: “Agents run your workflows on a schedule — each with a cadence, a status, and an execution history.” That’s the literal description Studio shows on the Agents list — it’s the most accurate one-line definition available, so this hub uses the same language the product does. Concretely, an Agent has:
  • Identity: title (display name), name (slug, set once at creation and immutable), description.
  • Status: active or inactive.
  • Frequency: how often it runs, stored as context["config-frequency"] in minutes (e.g. 0.5 = every 30 seconds).
  • Workflows: the one or more workflows it triggers on that cadence.
  • Documents: records associated with the agent, shown on its detail page.
  • Execution history: every past run, inspectable individually.

​
When to use an Agent

Use an Agent when you need something to run on its own, repeatedly — syncing data, refreshing embeddings, posting scheduled content — without a human or an external caller triggering it each time. If you need to run something on demand instead (a one-off pipeline triggered by a request), that’s a plain Workflow run directly, not an Agent.

​
The four things you’ll actually do

Creating Agents

How an Agent actually gets made today — Studio form vs. installing a skill/template.

Configuring Agents

The real fields you can set, and which ones the CLI can’t touch (yet).

Operating Agents

Running, watching, and inspecting executions from the terminal.

Agent Commands via CLI

A copy-paste command reference, plus CLI command vs. workflow vs. MCP tool.
  • Workflows — what an Agent actually triggers.
  • machina agent — the CLI command group.
  • MCP — how an Agent’s workflow can be exposed as a tool to an external agent.