Glossary of terms
The glossary provides terms and definitions for IBM® watsonx Orchestrate®. Whether you are a new user or an experienced developer, the glossary helps you understand the language and terminology that define the platform.
The following cross-references are used in the glossary:
-
See refers you from a nonpreferred term to the preferred term or from an abbreviation to the spelled-out form.
-
See also refers you to a related or contrasting term.
A B C D E F G H K L M P Q R S T U V W Z
A
action
A task or question that an assistant can help customers with. Each action has a beginning and an end, making up a conversation between the assistant and a customer.
action task
In a ruleflow, a task that contains rule action statements. These action statements are run each time the task is called.
agent
An AI-powered software entity that carries out tasks, makes decisions, and interacts with users or apps using models, tools, and workflows. Agents are designed for specific domains or business processes.
agent fleet
A collection of all agents associated with a tenant, managed as a single system for health, performance, and compliance.
agent fleet health
The overall operational condition of all your agents as a group. The agentic control plane provides administrators with centralized visibility into fleet-wide metrics.
agent memory
A capability that retains and recalls information about a user over time, across sessions and across agents. This information is used to personalize responses and improve the user experience.
agent-to-agent connection (A2A)
A connection that links one agent to another agent, including external or partner agents. It enables agents to communicate and delegate tasks to each other without human intervention.
agentic control plane
A centralized operational layer within watsonx Orchestrate that provides administrators with monitoring and management capabilities for AI agents deployed across enterprise environments. It enables administrators to monitor system health, investigate failures, manage compliance controls, and ensure agents operate reliably and securely throughout their lifecycle from design through deployment and ongoing operations.
AI-guided action
An action that allows a customer to choose an LLM, set a prompt, and optionally include a document.
app
A third-party application (for example, Salesforce, Workday, Onit) that connects to watsonx Orchestrate so agents and tools can perform actions such as retrieving data or triggering workflows.
asset
A reusable object that provides value within watsonx Orchestrate. Assets represent shareable capabilities used across the platform. Examples include agents and tools.
attributes and tags
Descriptive metadata associated with each span that includes details like the model in use, token usage, tool identifiers, API routes, response codes, and any custom notes.
automation
A business-defined process that requires minimal human input to complete tasks and fulfill a business need.
B
base LLM
A language model with a large number of parameters, trained on a large quantity of text.
bot
Attended or unattended software that automates repetitive tasks. Bots can be created to complete a wide variety of tasks, as well as manage and schedule the bots based on the nature of the task.
builder
A developer who develops apps and skills.
C
catalog
A centralized page where users can browse prebuilt agents, tools, and apps available in watsonx Orchestrate.
channels
Interfaces or platforms where an agent can be accessed or deployed, such as chats, embedded widgets, or integrated business applications.
choice list
A data type that contains a list of values.
composite data type
A data type composed of one or more attributes that can be primitive data types, simple data types, or other composite data types.
connector
The means by which a data source is connected to a product or service.
connection
A configuration that enables secure communication between watsonx Orchestrate and an external application. Connections define the authentication type, parameters, and environments for the integration.
content-based answering
A search integration method based on documents that a builder provides to complement an LLM's knowledge.
controls
Controls are system-level mechanisms within watsonx Orchestrate that allow administrators to establish and enforce rules for agents, models, tools, and system configurations. Two categories exist: Asset controls applies rules to AI components (models, agents, and MCP tools), while Enterprise controls manage tenant-wide settings for data retention, networking, and analytics.
custom-built action
An action that can be added to the assistant that is manually created, step-by-step, by the builder.
D
data type
A definition used to describe the structure of the data.
debug
To detect, diagnose, and eliminate errors.
decision
An skill type that captures and automates repeatable intelligent business decisions.
decision model
A model that contains a diagram and a decision logic.
draft environment
A safe, non‑production environment used for testing and validating connections, credentials, tools, and agent behavior before going live.
F
fallback
A conversation flow that end users can be routed to, and builders can modify the behavior of, in an assistant.
function task
In a ruleflow, a task that refers the output of another decision or prediction in the same automation.
G
general-purpose answering
A search integration method based on general knowledge rather than provided content, such as ChatGPT.
global workspace
A shared, permanent workspace available to all builders. All assets are visible to everyone, and access cannot be modified. Best suited for shared and reusable resources.
guardrails
Safety rules like PII redaction, allowed tools/models, and enforcement that protect against unintended behavior.
H
human task
A single unit of work generated when user input is required.
J
jump back in
A quick-return method for resuming recent work and navigating to frequently used agents.
K
knowledge
Structured or unstructured information used by an agent to reason, answer questions, or take action, such as domain knowledge, contextual data, or configured prompt templates.
L
live environment
The production environment where agents, tools, and connections run for real users and real business workflows.
M
MCP server
A server that provides agents with access to external tools and data through a standardized protocol. It enables agents to discover and use these capabilities while performing tasks. Ideal for integrating agents with external systems that support MCP.
member credentials
Personal credentials tied to an individual user.
N
needs attention
A prioritized status indicator that highlights both immediate high-priority operational issues requiring action and proactive insights designed to prevent future problems. This helps administrators identify and take appropriate actions to resolve issues.
P
parent and child spans
A structured hierarchy that illustrates how operations are connected. It shows which actions triggered others and how tasks are nested within a larger workflow.
prediction model
A model that implements a prediction based on artificial intelligence. A prediction returns a probability or score that can be used in other decisions and workflows.
preview
To test and validate an automation during the build phase.
primitive data type
A set of basic data types from which all other data types are constructed.
private workspace
A restricted workspace for project-specific work. Builders or admins can create it, owners manage member access, and only members can view or edit assets. Ideal for team projects, prototypes, and sensitive work.
project
A workspace for building and managing skills like workflows, decisions, generative AI, connectors, skill flows and others.
prompt
Data, such as text or an image, that prepares, instructs, or conditions a foundation model's output.
publish
To make the public skills within a project available in the skill catalog for others to use. For example, to publish a skill so that it can be used by an assistant or by other skills.
R
ruleflow model
A model that contains rule artifacts and a ruleflow to organize their execution.
rule task
In a ruleflow, a task that lists rule artifacts in the order in which they need to be run.
S
SaaS
service user
In SaaS, a user who is a member of the client account.
simple data type
A data type that is an extension of a primitive data type. For example, an email type is an extension of the string primitive with a constraint value.
skill-based action
An action that can be added to the assistant that uses prebuilt skills that automate tasks for users.
software as a service (SaaS)
A model of software deployment whereby software including business processes, enterprise applications, and collaboration tools, are provided as a service to customers through the cloud.
span
An individual unit of work within the system, such as an LLM request, a tool call, or a database query. Each span records timing details, duration, and related metadata.
subflow task
In a ruleflow, a task that refers another ruleflow. A subflow task can refer to a ruleflow in the current ruleflow model, or in another ruleflow model in the same project.
T
task
A unit of work performed by someone or something.
team credentials
Shared credentials that allow all team members to access an application through a connection.
tenant
An entity that accesses services through a subscription associated with an account or namespace.
tool
A reusable action that integrates directly with an external application to perform a specific function (such as "Create legal matter" or "List invoices"). Tools act as building blocks used by agents.
toolset
A collection of tools required by an agent to perform actions in an external system. Each agent lists the specific toolset it depends on.
trace
An end-to-end record of a single user interaction, starting from the initial request and continuing through every processing stage until the final output. It captures the full execution lifecycle.
V
version
The result of capturing the state of a project at a specific point in time.
W
workflow
A skill type that is composed of a sequence of skills and tasks that define work.
workspace
A workspace is an isolated environment in watsonx Orchestrate that is used to organize, develop, and manage agents and related artifacts. Workspaces enable controlled access, collaboration, and separation between development and production activities.
workspace artifacts
Workspace artifacts are resources created and managed within a workspace, including agents, tools, and knowledge bases. Artifacts are scoped to the workspace during development.
workspace member
A workspace member is a user who has access to a workspace. Members are assigned roles which determine the actions they can perform on workspace artifacts.