Key concepts

Klu organizes the AI application lifecycle into a small set of related objects. Learn these terms before you follow the guides or use the API.

Product hierarchy

Workspace

A Workspace is the top-level boundary for your team. It owns Apps and shares members, model providers, Connections, Context, Skills, API keys, and settings.

You usually use one Workspace for one organization or team. Access and feature availability can depend on your Workspace plan and configuration.

App

An App groups related Actions and their operational data. Inside an App, the current UI gives you access to Studio, Datasets, Logs, Curators, Evaluate, and Finetune.

Use separate Apps when features have different owners, data, or release lifecycles.

Action

An Action is reusable AI functionality. It combines a prompt, model configuration, variables, optional Context and Skills, and output settings.

The current Action types are:

• Prompt for a completion-style request.
• Chat for a message-based assistant.
• Workflow for a sequence of connected steps.

An Action can have multiple versions. Publish and deploy the version that your application should call.

Studio

Studio is where you create, test, compare, and save Actions. You select a provider and model, edit the prompt, supply variable values, attach Context or Skills, and inspect the result before deployment.

Connections and grounding

Model provider

A model provider gives Klu access to one or more models. You add provider credentials in Connections, and Actions refer to the resulting Workspace provider configuration.

Provider and model availability can differ by Workspace. Use the selectors in Connections and Studio as the current source of truth for your account.

Connection

A Connection authorizes Klu to access a model provider or third-party service. Connections are shared at the Workspace level. Removing or changing a Connection can affect every Action, Context source, or Skill that depends on it.

Context

Context is a searchable collection of information used for retrieval-augmented generation. You add documents or connect a source, then attach the Context to an Action so relevant passages can be included at inference time.

Retrieval quality depends on the source material and the query. Context can reduce unsupported answers, but it does not guarantee factual output.

Skill

A Skill lets an Action invoke a configured API or computation. Use a Skill when the model needs to perform an operation; use Context when it needs to retrieve information.

Logs, Datasets, and improvement

Log

A Log is a recorded Action generation and its operational metadata. Logs can include the input, output, model information, latency, user or source metadata, and feedback.

Review Logs to investigate behavior and identify examples worth adding to a Dataset. Avoid sending sensitive metadata unless your application and Workspace policies allow it.

Feedback

Feedback is a quality signal attached to a Log, such as a positive or negative rating, correction, issue, action, or comment. Feedback is evidence about one interaction; aggregate it before drawing conclusions about overall quality.

Dataset

A Dataset is a curated collection of examples. You can build one from Logs or import data, then use it for evaluations, export, or supported fine-tuning workflows.

Keep evaluation examples representative and separate from examples used to tune the same behavior. This helps you avoid measuring memorization as improvement.

Evaluation

An evaluation measures an Action or model against a Dataset and configured evaluators. Compare runs using the same test cases and criteria. A higher score is meaningful only when the evaluator and Dataset match your production goal.

Insights and Finetune

Insights summarizes performance, feedback, and usage across the Workspace. Finetune trains a supported provider model from an eligible Dataset. Fine-tuning availability and minimum data requirements depend on the selected provider and Workspace configuration.

UI and API boundaries

Use the Klu app to configure Workspaces, Apps, Actions, providers, Context, Skills, Datasets, and evaluations through the UI.

Use the Klu API reference or the Python and TypeScript SDKs to manage supported resources and call deployed Actions from your code. The UI can expose configuration that is not represented by every SDK helper, so use the API reference for the current request and response schema.