Skip to main content

Glossary

The Glossary section provides clear definitions of important terms, technologies, and concepts used across the Loc.ai platform and its documentation. It serves as a quick reference guide to help users understand technical language, system components, and product-specific terminology.

Entries are organised alphabetically for easy navigation and faster lookup. This section is especially useful for new users, administrators, developers, and teams working with AI-powered location intelligence solutions.

1. Agent​

The Loc.ai software running on an edge device. The agent handles local model execution, command polling, hardware sensor interfacing, and telemetry reporting back to the Control platform. Sometimes used interchangeably with Loc.ai.

2. API key​

A unique token issued to each device upon activation. Sent in the Authorization: Bearer header on every request to authenticate the device with the Control platform. API keys are device-scoped a key issued to one device cannot be used to access or impersonate another. Keys can be revoked from the dashboard at any time.

3. Canary deployment​

A deployment strategy where a model update is rolled out to a small subset of devices first. Results and stability are validated before the deployment is widened to the rest of the fleet. Useful for catching regressions without fleet-wide disruption.

4. Control plane​

The centralised backend service (Loc.ai:Control) responsible for managing device lifecycles, orchestrating deployments, processing telemetry, and storing inference results. Accessed via the web dashboard or API.

5. Device​

Any machine registered with the Loc.ai platform and running Loc.ai. Supported hardware includes Windows, macOS, and Linux PCs, local servers, and edge hardware such as NVIDIA Jetson and Intel NUC. Each registered device has a unique API key and appears in the Control platform dashboard.

6. Device token​

A one-time registration key generated by the account owner and used to activate a new device. Once used, the token is invalidated and replaced by a persistent API key for ongoing authentication. Also referred to as a registration key.

7. Edge​

Computing that occurs close to the data source on-premises, on IoT hardware, or on end-user machines rather than in a centralised cloud data centre. Loc.ai's core value proposition is moving AI inference to the edge to reduce latency, cost, and data egress.

8. Edge function​

Serverless-style logic executed on an edge device to process data locally before or after model inference. Enables lightweight pre- and post-processing without round-tripping to the cloud.

9. Fleet​

The complete collection of devices registered under a Loc.ai account. Fleets are typically organised using tags (e.g. by location, environment, or hardware type) to enable targeted deployments and filtered analytics.

10. GGUF​

A model serialisation format used by llama-cpp for efficient language model inference on CPU hardware. Loc.ai supports GGUF models for on-device LLM tasks. GGUF replaces the older GGML format and is the current standard for llama-cpp-compatible models.

11. Inference​

The process of running a trained model against an input to produce an output a classification label, a confidence score, or a generated text response. In Loc.ai, inference runs entirely on the device via Loc.ai. Results are returned to the calling application synchronously and reported to the Control platform asynchronously.

12. Llama-cpp​

An open-source inference runtime optimised for running GGUF language models on CPU hardware, without requiring a GPU. Loc.ai uses llama-cpp as the execution backend for all LLM inference tasks.

13. Loc.ai:Control​

The central management platform, consisting of a FastAPI backend and a React web dashboard. Used to register devices, manage fleets, deploy models, monitor telemetry, and view inference results.

14. Loc.ai:Link​

Loc.ai:Link enables on-device AI inference with local sensor processing and real-time edge intelligence.It seamlessly connects devices to secure cloud orchestration for scalable, managed AI operations.

15. Loc.ai​

The lightweight Python agent installed on each registered device. Loc.ai handles device authentication, model downloads, local inference execution via TFLite or llama-cpp, and telemetry reporting. Communicates with the Control platform over HTTPS using poll-based command retrieval no inbound ports are required on the device.

16. Model​

A trained machine learning artifact deployed to edge devices for local inference. Models are stored and versioned in the Model Library. Supported formats are TFLite (for image and audio classification) and GGUF (for language models).

17. Model registry​

The central repository within the Loc.ai platform where model artifacts are uploaded, versioned, and managed before being targeted to devices via deployments. Also referred to as the Model Library.

18. OTA update​

Over-the-air update the delivery of new software, model versions, or runtime configuration to a device without physical access. In Loc.ai, OTA updates are initiated from the Control platform and delivered to devices via signed artifact URLs fetched on the next command poll.

19. Registration key​

See device token a one-time key used to register a new device with the platform during initial setup.

20. Rolling deployment​

A deployment strategy where updates are applied to devices one at a time, or in small batches, to minimise fleet-wide disruption. Contrasted with a simultaneous push to all devices.

21. Tag​

A label applied to devices for organisational and targeting purposes. Tags can represent location, environment (e.g. production, staging), hardware type, or any other grouping relevant to your fleet. Deployments can be scoped to devices matching specific tags.

22. Telemetry​

Structured health and performance data reported by Loc.ai to the Control platform at regular intervals (every 30 seconds by default). Telemetry includes CPU utilisation, RAM usage, temperature, GPU load, and inference statistics. Raw input data and model outputs are not included in telemetry payloads.

23. TFLite​

TensorFlow Lite a model format and inference runtime optimised for on-device execution on resource-constrained hardware. Loc.ai uses TFLite to run image classification and audio classification models on registered devices.