Concepts

The platform is built from a small set of objects. This section has a page per object, each answering the same four questions: what it is, why it exists, what it relates to, and where you see it in the web application, the CLI and through Claude.

Read them in this order the first time. Each page assumes the ones above it.

Start here

1. Organization and project: who your account belongs to, and the box that holds one piece of work.
2. Device: a machine running the CLIKA agent, what it reports, and everything you can do with one.

Runtime

What you do with a model once the platform can reach your hardware.

3. Benchmark: a run over models, tests and devices, and every benchmark in the catalogue.
4. Model deployment: one of your models running on your devices behind an endpoint the platform proxies.
5. Runtime licenses: the credential a ClikaRT runtime presents, its entitlements, and how you issue and revoke one.

Device management

The machinery underneath, and what you reach for when the built-in paths are not enough.

6. Artifact: the versioned files the platform stores and puts on devices, including external Docker and Git sources.
7. Job: the single execution of one benchmark on one device, and its job-definition template.
8. Service: a process you tell the platform to keep running on a device.

How they fit together

Organization                     one company or team; holds users and roles
 └── Project                     one piece of work; holds its own resources
      ├── Model                  a Hugging Face repository plus its task
      ├── Benchmark run          models x tests x devices, one result set
      │    └── Job               one model on one device, one execution
      │         └── Result       quality scores, performance metrics, per-item data
      ├── Model deployment       a model served on devices, behind the platform proxy
      ├── Service                a process kept running on a device
      ├── Artifact               a versioned file the platform puts on devices
      └── Runtime license        what a ClikaRT runtime presents to run

Devices belong to the organization, not to a project, so every project of an
organization sees the same fleet.

The direction of causality is worth holding on to. A benchmark run is what you create; the platform fans it out into one job per model and device pair, each job runs a job definition the platform picked for that device, and each job produces one result. You never create a job by hand for an ordinary benchmark, and you never pick the job definition.

The vocabulary that trips people up

Term	What it means here
Benchmark	One run you created, over one or more models, tests and devices. The API calls it a benchmark group.
Quality test	One named test inside a run (MMLU, GSM8K, IFEval). The API calls it a benchmark type.
Job	One execution of one model on one device inside a run.
Job definition	The template a job executes: artifacts, setup, script, teardown, output path.
Deployment (in the web application)	One of your models served on your devices. See Model deployment.
Serving	The API's name for the same object.
Runtime license	The credential a ClikaRT runtime presents. Issued by your platform, per project. See Runtime licenses.
Platform license	A different artifact: what CLIKA issues to activate one deployment of the platform itself. See Activate a deployment license.

Where to go next

• Getting started if you have not run anything yet.
• How-to guides once the vocabulary is familiar.