Glossary
Quick reference for terms used throughout qRaptor documentation.
A
A2A Protocol
Agent2Agent — an open protocol for one agent to call another over JSON-RPC 2.0. Lets you expose an AI Agent as a service that other agents or external systems can call programmatically. qRaptor supports A2A 0.2.x and 0.3.0 via the message/send method. See A2A integration.
AG-UI Protocol
The streaming protocol that powers qRaptor’s conversational interface. Delivers real-time thinking, planning, and response updates as they happen.
Agent
An AI-powered execution unit built on qRaptor. See AI Agent and Task Agent.
Agent Graph
A MAS pattern where you draw the flow yourself: agents as steps, connections between them, conditions on the connections, and loops where work needs another pass. See Agent Graph overview.
Agent Persona
A configuration that defines an agent’s voice, personality, communication style, and behavioral guidelines.
Agentic Workspace
The section of a project where you build, test, and manage agents, MAS systems, personas, and agent tools.
AI Agent
A conversational agent that reasons autonomously, uses tools, accesses memory, and responds in natural language. Built for interactive conversations and complex decision-making.
AI Assist
The button inside a skill’s System Prompt editor that drafts or refines that one prompt. Distinct from AI Generate, which builds whole skills. See Advanced skill settings.
AI Generate
The feature that drafts complete skills for an AI Agent — name, description, tags, system prompt and tools — from a short description plus the tools, tables, documents and content stores you select. It can also refine an existing skill in place. See Generate & refine skills with AI.
Context Graph
A map of an existing codebase plus a semantic search over it, built by ingesting its repositories. Lets an agent understand code it did not write. Distinct from the App Context Graph, which describes a generated app. See Context Graph.
Context Graph application
One system in the Context Graph, holding every repository that belongs to it. Cross-repo links only exist inside one application. See Create an application.
App Context Graph
A structured representation of a generated application’s entities, files, and relationships. Powers Vibe Code.
Appliance
A project exported as a sealed, signed archive that runs on infrastructure you control, carrying the app, its agents, the governed data model, the policies and the runtime that enforces them. Once installed it makes no runtime calls to qRaptor. See Export as an appliance.
C
Canvas
The visual drag-and-drop graph editor where you design agent execution flows by connecting nodes.
Channel
A communication endpoint where end-users interact with your deployed agents. Supported: Web, WhatsApp, Voice, SMS, API.
Content Store
A vector store for RAG (Retrieval-Augmented Generation). Upload documents, and your agents can semantically search them for relevant knowledge.
Code Workspace
A configuration saying where an agent’s copy of a repository lives, which branch it tracks, and which files and branches it may never touch. One per project. See Code & Execution.
Command Sandbox
A configuration saying how an agent’s commands run and which are permitted. Starts as disabled, so nothing runs until it is set. See Code & Execution.
Conversation Router
A MAS pattern where each turn is answered by one specialist agent, chosen by classifying the message against every member’s skills. See MAS overview.
Copilot
An embeddable AI chat widget that connects to your agents. Deploy as a web component on any site.
Credit
The unit of consumption for LLM calls and agent executions. Your plan determines your monthly credit allowance.
D
Data Vault
The secure data layer within a project: Tables, Documents, Content Stores, and Memory.
Deployment
A published instance of your agents or apps serving traffic. See Preview and Production.
Document
A file stored in the Data Vault. Supports uploads, exports, and attachments.
E
Episodic Memory
One of three memory tiers. Records what happened in past sessions — session summaries that the agent can recall in future conversations.
Enrichment
An optional ingest stage where a model proposes what the parsers cannot determine — a type, a description, an authorization pattern. Off by default, and everything it proposes is reviewed before it counts. See Review enrichment.
Executor
See Remote Executor.
G
Graph
The directed flow of nodes that defines an agent’s execution logic. Built visually on the canvas.
H
Graph version
The immutable result of one ingest, pinning one commit per repository. Agents cite it, which is why it cannot change afterwards. See Versions & diffs.
Human Gate
A node type that pauses agent execution for human approval. Supports synchronous (wait for immediate response) and asynchronous (queue for later review) modes. Also called Human-in-the-Loop (HIL).
I
Ingest
A job that reads a Context Graph application’s repositories and builds a graph and search index from them. Each completed ingest writes one graph version. See Run an ingest.
Integration
A connection to an external service — API, webhook, or MCP server — that extends your agents’ capabilities.
L
LLM Task Node
A node that sends a prompt to a large language model with full configuration: persona, tools, policies, structured output, and memory access.
M
MAS (Multi-Agent System)
A system that coordinates multiple agents working together using an orchestration pattern.
MCP (Model Context Protocol)
An open standard for connecting AI systems to external tools and data sources. qRaptor connects to MCP servers to give your agents access to external tools and data.
Memory
Persistent storage that agents use to remember information. Three tiers:
N
Node
A single step in a Task Agent’s graph. qRaptor provides node types across categories: Flow Control, Logic, Loops, Parallel, Actions, Data, Memory, RAG, and Code.
O
Orchestration Pattern
The coordination strategy for a MAS, chosen when the system is created. There are three: Conversation Router, Planner / Executor and Agent Graph. See MAS overview.
Output key
Where an agent’s result is stored in an Agent Graph’s shared scratchpad. Later agents read it, and conditions test it, as stm.<key>. See Agents on the canvas.
P
Plan
The pricing tier for your workspace. Tiers: Free, App Lite, Starter, Pro, Team, Enterprise.
Planner / Executor
A MAS pattern where a planner reads the goal and writes a plan — which agents run, in what order, and what runs at the same time — and an executor runs it. See Planner / Executor overview.
Preview
A deployment environment for safe testing. Only your team can access it.
Production
The live deployment environment serving real end-users at yourapp.qraptor.app.
Profile Memory
One of three memory tiers. Stores end-user preferences and attributes (explicit and LLM-inferred) that persist across conversations.
Project
A container that groups agents, apps, data, tools, and deployments for a single product or initiative.
Q
Quality Gate
A configuration holding your definition of done — whether the build must pass, how many failing tests are tolerable, the minimum coverage. One tool applies all of it and returns a single verdict. See Code & Execution.
Quick Prompt Node
A simplified LLM node for one-shot prompts without persona, memory, or tool access.
R
RAG (Retrieval-Augmented Generation)
A technique where the agent searches relevant documents from a Content Store before generating a response, grounding answers in your data.
RBAC (Role-Based Access Control)
The permission model controlling who can do what. Roles: Owner, Admin, Developer, Viewer.
Remote Executor
A lightweight agent runtime you deploy in your own infrastructure (on-premise, VPC, or GPU). Connects to qRaptor via secure WebSocket. Lets agents run close to your data.
Run
A single execution of an agent from trigger to completion. Tracked with timing, cost, node-by-node progress, and trace data.
S
Semantic Memory
One of three memory tiers. Facts, rules, and knowledge the agent learns from its executions — shared across all users of that agent.
Session
A conversation context between an end-user and an agent. Sessions maintain state across multiple messages within a single interaction.
Skill
A named capability of an AI Agent, with its own description, tags, system prompt and optional tool list. An agent matches each incoming message against its skills and applies the best fit. Skills are also what a MAS router uses to decide which agent answers. See Skills overview.
Step
One agent placed on an Agent Graph canvas, or one entry in a plan produced by Planner / Executor.
Studio
The qRaptor web application at studio.qraptor.ai where you build, test, and manage everything.
T
Table
A structured data store in the Data Vault. Supports schema design, ER diagrams, and a SQL query explorer.
Task Agent
A workflow-based agent that executes a defined graph of steps. Nodes run in sequence with branching, loops, and parallel execution. Ideal for deterministic pipelines and automations.
Tool
An external capability an agent can invoke: API calls, database queries, code execution, file operations, web search, and more.
Toolchain
A configuration holding the commands that build, test and check one repository. This is what lets an agent work in any language without a platform change. See Code & Execution.
Trace
A complete record of a request flowing through the system. Shows timing, node execution, tool calls, and any errors.
V
Version
An immutable snapshot of an agent’s graph, configuration, and tools. Created automatically on save. You choose which version to deploy.
Vibe Code
The AI-powered code generation system that turns natural language prompts into full-stack applications through the pipeline: Prompt → Requirements → App Graph → Code → Auto-Fix → Iterate.
W
Workspace
Your top-level account and the billing & isolation boundary — everything you build lives in one workspace, fully isolated from other workspaces. Each workspace is on a plan. (Not to be confused with the Agentic Workspace, the section inside a project where you build agents, or the app workspace used for generated applications.)
Can’t find a term? Contact us at innovate@augmentappz.ai.