Glossary

Definitions of key terms and concepts used throughout the Loc.ai platform and documentation.

A

Agent

The Loc.ai:Link software on edge devices. Handles local model execution, control plane communication, and telemetry.

API key

A unique token issued to each device upon activation. Authenticates requests between the agent and the control plane.

C

Control plane

The centralised backend (Loc.ai:Control) that manages devices, orchestrates deployments, and processes telemetry.

Canary deployment

Rollout to a small percentage of devices first to validate stability before wider release.

D

Device

Edge hardware running Loc.ai:Link (e.g. NVIDIA Jetson, Intel NUC, custom x86_64).

Device token

One-time registration key linking a device to your Loc.ai account during setup.

E

Edge

Computing close to data sources (on-premises, IoT) rather than only in centralized clouds.

Edge function

Serverless-style code on edge devices to process data locally.

F

Fleet

A collection of edge devices managed through Loc.ai:Control, often organised with tags.

I

Inference

Running a trained model to produce outputs from inputs. In Loc.ai, inference runs locally on edge devices.

M

Model

A trained ML artifact (LLM, vision, etc.) deployed to edge devices for local inference.

Model registry

Central repository where model artifacts are stored and versioned before deployment.

O

OTA update

Over-the-air update of software or configuration without physical access.

R

Registration key

See device token — one-time key to register a device with the platform.

Rolling deployment

Updates applied one device or small batch at a time to minimise disruption.

T

Tag

Label on devices for organisation and targeting (location, environment, etc.).

Telemetry

Health and performance data from devices: CPU, memory, GPU, inference stats.