---
title: "Machines & Fleet"
canonical: https://trickest.com/docs/key-concepts/machines-and-fleet
description: "Distinguish workflow nodes, execution tasks, fleet capacity, and agent sandboxes."
---

# Machines & Fleet

## What are Machines and Fleets

A **fleet** is a pool of machines available for workflow execution. You choose the fleet when starting a run. A **machine** supplies the compute for an execution job.

These terms describe different parts of a run:

| Term | Meaning |
|---|---|
| Node | A step in the workflow definition. |
| Task / job | An execution of that step with particular inputs. A distributed node can have many tasks. |
| Machine | A worker that executes a job. |
| Fleet | The pool of machines selected for a run. |

A workflow with 500 tasks does not need 500 machines at once. Available capacity, the run's machine limit, and dependencies determine how much can run concurrently.

## Fleet Types

| | Managed | Self-hosted |
|---|---|---|
| Who operates the machines | Trickest | You |
| Setup | Select an available managed fleet | Attach and maintain your hosts |
| Capacity | Subject to the fleet and run limits | Subject to limits and the hosts you keep online |
| Typical use | Execution without managing worker infrastructure | Internal network access or your own infrastructure |

Managed workflow compute is an Enterprise capability. Community workflows use self-hosted machines. Check your plan and **Settings > Infrastructure > Fleet** for the capacity available to your account.

## How a Run Uses a Fleet

The execution engine assigns ready jobs to available workers. Dependent jobs wait for their inputs; jobs can also queue when capacity is occupied. Multiple runs can share a fleet, so a machine limit is an upper bound rather than a reservation.

Parallel branches and [distributed nodes](/docs/using-the-app/workflow-and-executions/distributing-and-scaling-jobs) can use more workers, but increasing the limit does not create parallel work in a sequential graph.

## Workflow machines versus agent sandboxes

An [agent sandbox](/docs/using-the-app/ai-agent/sandboxes) is the agent's working environment for commands, files, and app previews. It has a separate lifecycle from workflow execution. A running agent sandbox does not prove that the selected fleet has an online machine for a workflow run.

## Next Steps

- [Managed Machines](/docs/key-concepts/managed-machines): capacity and shared fleet behavior.
- [Self-Hosted Machines](/docs/key-concepts/self-hosted-machines): host responsibilities and connectivity.
- [Distributing and Scaling Jobs](/docs/using-the-app/workflow-and-executions/distributing-and-scaling-jobs): split work across tasks.

---
_Markdown source of https://trickest.com/docs/key-concepts/machines-and-fleet._