Skip to main content

Appendix D: Glossary of Terms

Agent​

An LLM that has been configured with a specific persona, a set of rules, and a curated list of Tools. An Agent's primary role is to receive a high-level goal and create a plan (a DAG) to achieve it.

Pack​

A self-contained, installable unit of functionality for Worka. A pack can contain A2UI views, MCP tools, schemas, SQL migrations, and assets. It is the fundamental unit of distribution in the ecosystem.

Capability​

A specific permission that a pack must declare in its aip.json manifest to use a protected feature of the Host application (e.g., net.access).

Conversation​

A complete, long-running record of a single workflow from the initial prompt to the final result. It is identified by a conversation_id and contains all associated events, steps, and logs.

DAG (Directed Acyclic Graph)​

The plan an Agent creates to achieve a goal. A DAG is a flowchart of steps where each step is a Tool Call, and the graph defines the order of execution and the flow of data between them.

Elicitation​

The protocol that allows a backend MCP Server to pause its execution and send a request back to the Worka Host. This is used to either request input from the user or to ask the Host to call a tool on another pack.

Host Application​

The main, native Worka desktop application that the user installs. It is responsible for running and managing all AI Packs, providing core services, and ensuring security.

MCP (Model Context Protocol)​

The JSON-based communication standard that all backend servers must implement. It defines the format for requests (like tool.call) and responses, allowing the Host to communicate with backends written in any language.

MCP Server​

The backend implementation of a pack’s tools. In Worka v2 it runs in‑process (WASM) and communicates via JSON MCP requests handled by the host.

Orchestrator​

The core AI engine built into the Worka Host. It is responsible for taking the DAGs generated by Agents and executing them step-by-step.

Strategy​

The set of instructions provided to an Agent to guide its behavior and planning process. A strategy is typically composed of a natural language description (the persona) and a set of structured rules.

Tool​

A single, discrete function exposed by an MCP Server. A Tool is the bridge between the AI's reasoning and the real world, allowing it to perform actions like searching the web, reading a file, or sending an email.

Virtual Pack​

A special, built-in pack that does not exist as files on disk but instead provides a direct API to core Host functions. Examples include worka/ui, worka/db, and worka/orchestrator.

Worka SDK​

Language‑specific SDKs (Rust, Swift, Java, etc.) that provide typed builders for A2UI and helpers for MCP tooling. They generate A2UI JSON without requiring a web runtime.