Glossary

Key terms and concepts used in Fabric

A

Agent

The Fabric Agent is the desktop application that device owners install to participate in the network. It consists of a UI component and a background daemon.


D

Device Owner

Someone who contributes their computer idle resources to the Fabric network in exchange for earnings. Also called a Node Provider.


E

Enrollment

The process of registering your device with the Fabric network. During enrollment, your hardware is detected and a unique device ID is generated.


H

Heartbeat

A periodic signal sent by your device to the backend to indicate it is online and available. Missed heartbeats cause your node to show as offline.


I

Inference

Using a trained AI model to make predictions. For example, running an image through an image classifier to identify what is in it.


J

Job

A unit of work submitted to the Fabric network. Jobs have a type (e.g., text_summarization), parameters, and produce results.


M

ML Pipeline

A series of machine learning steps: preprocessing data, training a model, and running inference. Fabric supports end-to-end ML pipelines.


N

Node

A device participating in the Fabric network. Each enrolled device becomes a node that can execute workloads.

Node Provider

See Device Owner.


O

Overage

Usage beyond your subscription plan included limits. Overages are billed at per-unit rates.


P

Payout

Money transferred to device owners for completed work. Payouts are processed via Stripe Connect.

Proof of Execution (PoE)

A cryptographic record generated after each job completion. Contains hashes of workload code, input, and output to verify the work was done correctly.


S

Sandbox

An isolated execution environment that restricts what workloads can access. Prevents malicious code from affecting the host system.

Stripe Connect

The payment platform used to send payouts to device owners. Requires identity verification and bank account connection.


T

Training

The process of teaching a machine learning model to make predictions by showing it examples. Fabric supports distributed training workloads.


U

Unenroll

The process of removing your device from the Fabric network. After unenrolling, your device will no longer receive jobs.


W

WebSocket

A communication protocol that enables real-time, bidirectional communication. Fabric uses WebSockets for instant job notifications.

Widget

The embeddable JavaScript component that adds an AI Assistant chat interface to your website.

Workload

A specific type of computation that can be executed on the network. Examples: text_summarization, video_transcode, ml_train.