Glossary
This glossary defines terms used throughout the platform. Where the current UI uses a different name than older documentation, the current name is listed first.
Agent — A configurable virtual assistant that combines instructions, an AI model, and optional toolkits to perform tasks and answer questions.
AI Providers — The Settings section where you configure LLM, embedding, image generation, speech recognition, and text-to-speech models and their credentials for a project.
Artifact — A file stored in a project bucket, shared between agents, pipelines, and conversations for temporary storage and context.
Bucket — A container in Artifacts that groups related files under a single retention policy.
Canvas — The in-chat editor for code, table, and Mermaid-diagram blocks generated during a conversation. It edits the block created in the current turn; it does not open or save previously stored files.
Catalog (Elitea Catalog) — The shared library of published agents and skills, browsable and searchable across all your projects. Older documentation calls this “Agent HUB” or “Agents Studio.”
Chat — The conversational interface where you interact with models, agents, pipelines, toolkits, MCP servers, and teammates.
Context management — Automated pruning and summarization of older conversation messages so long conversations stay within a model’s token limit. Configured in Settings › Memory.
Conversation Starter — A predefined prompt configured on an agent or pipeline that helps users begin an interaction.
Credential — Securely stored authentication information (API key, token, or similar) used by a toolkit or AI provider to connect to an external service.
Entity — General term for the main things you create and manage: agents, pipelines, skills, toolkits, and artifacts.
Indexing — The process of extracting content from files and toolkits and storing it in a vector database so it can be retrieved with semantic search.
LLM (Large Language Model) — The AI model that processes and generates text for a conversation, agent, or pipeline.
MCP (Model Context Protocol) — A protocol that lets Elitea connect to external MCP servers over HTTP, exposing their tools to agents, pipelines, and conversations. Elitea itself also acts as an MCP server, exposing agents (tagged mcp) and MCP-enabled toolkit tools to external MCP clients.
Memory — In Settings, the personal page that configures context-window management (pruning, summarization) for new conversations. Persistent, cross-conversation long-term memory is not available yet.
Node — A single step in a pipeline (agent, toolkit, MCP, LLM, decision, router, state modifier, printer, or human-in-the-loop) that the pipeline compiler can run.
Participant — Anything active in a conversation: a model, an agent, a pipeline, a toolkit, an MCP server, or a human teammate.
Personal Access Token (PAT) — A token you generate in Settings › Personal Tokens for programmatic API access or IDE integration.
Pipeline — A visual, multi-step workflow connecting agents, toolkits, and control-flow nodes into a repeatable automated process.
Project Context — A project-level background document that Elitea applies to every AI interaction in the project, giving agents shared domain terminology, workflows, and constraints.
Public Project — A shared, deployment-wide project (separate from your personal and team projects) that some settings sections — Project Context, Prompts, Environment — are scoped to.
Remote MCP — An MCP connection to an HTTP/HTTPS-based server, configured like any other toolkit type.
Skill — A reusable, Markdown-based instruction set attached to an agent version and invoked on demand with ~skill-name.
Support Assistant — An AI-powered help widget, embedded in the platform, that is aware of the project and page you are on.
Tag — A label used to categorize and filter agents, pipelines, and other entities.
Toolkit — A configured integration — with an external service or an Elitea-native capability — that gives agents and pipelines the ability to act. See the toolkit catalogue.
Token — A unit of text used by AI models to measure input and output size, and also the name for a personal access token depending on context.
Version — A saved snapshot of an agent, pipeline, or skill. Agents and pipelines support multiple named versions; skills currently support only one implicit base version.
Vector Storage — The storage layer used for embeddings and semantic retrieval: Qdrant, behind the elitea-vector service. It is set up once for the whole deployment by an administrator; projects and toolkits configure nothing, and each project’s vectors are kept apart.
Webhook — A callback-based mechanism some pipeline triggers use to start a run from an external event. Webhook management is available through the API only; there is no management page in the UI yet.
Terms you may see in older material
Section titled “Terms you may see in older material”These names appear in earlier documentation but describe something that has changed shape or is no longer part of the platform. They are listed here so the term is not a dead end.
Agent HUB / Agents Studio → Elitea Catalog
The shared library of published agents and skills is now the Catalog tab, reachable from the sidebar.
Local MCP / Elitea MCP Client / STDIO transport
Elitea’s MCP server is streamable-HTTP only. There is no local STDIO MCP client or server; only Remote MCP connections exist.
Elitea Code (IDE extensions)
No trace of dedicated VS Code or IntelliJ extensions exists in the current platform. Personal tokens still expose IDE settings snippets on the Personal Tokens page.
Langfuse
Tracing is provided through OpenTelemetry via the platform’s own collector; there is no Langfuse integration.
ReAct / XMLChat agent types
These were older application types. The current agent-type values are openai (default), react, dial, and pipeline.
Top K / Top P sampling parameters
Not exposed in the current model-settings UI. Use Temperature or Creativity/Reasoning instead, depending on the model.
Collection
“Collection” as a distinct entity type is not present in the current platform; content is organized with tags and folders instead.