OCI Digital Assistant Professional key terms

The top-level conversational container that groups related skills, routes user requests between them, and exposes a single interface to channels. It holds shared configuration such as the default language and the confidence threshold used for routing.

A self-contained conversational unit inside a digital assistant that handles one topic area, with its own dialog flow, intents, entities, and backend integrations. A skill can also be published standalone as a skill bot.

The browser canvas where you build a skill's dialog as a graph of states and transitions. It offers a visual mode and a YAML code mode that edit the same underlying flow definition.

The ordered set of states that controls how a skill responds to user input, from the entry state through resolution, confirmation, and completion. It defines what the assistant says and does at each turn.

A single node in a dialog flow that performs one job, such as sending a message, collecting input, evaluating a condition, invoking a component, or ending the conversation. Each state has a type and configuration.

A rule that determines which state runs next after the current one finishes. Transitions can be unconditional, action-based, or conditional on variable values, and may set variables when taken.

A named category representing what the user wants to accomplish, such as ordering a coffee or checking a balance. The intent is detected from user utterances so the skill can pick the right flow.

An example phrase a user might say to express an intent, used to train intent classification. Good utterance sets cover varied wording, synonyms, and entity placeholders rather than one fixed sentence.

A structured piece of information extracted from user input, such as a date, city, or product name. Entities can be built-in, custom lists, or composite, and resolved values are stored in variables.

A custom entity whose items are themselves entities, letting you capture several related values in one turn, such as a full address. It supports prompts, validation, and partial extraction across turns.

A resource that normalizes a matched entity value into a canonical form, for example mapping "NYC" to "New York". Resolvers can use lists, regular expressions, or custom component logic.

The routing logic that decides which skill and intent best match a user utterance, using confidence scores and configured thresholds. It can fall back to an unresolved-intent handler when nothing matches well.

A configurable score boundary that determines whether a detected intent is accepted, clarified, or rejected. Setting it too low causes wrong matches; too high causes frequent fallbacks and re-prompts.

The handler that runs when no intent reaches the confidence threshold. It typically asks the user to rephrase, offers suggestions, or escalates to a human agent instead of failing silently.

A named value scoped to a skill or flow that stores user input, component output, or context. Variables can be typed, given default values, and referenced in messages, conditions, and transitions.

A variable whose value persists across the conversation rather than only within one flow, used to remember user preferences or session facts. Scope must be chosen carefully to avoid leaking data between users.

A built-in dialog building block provided by the platform, such as a text-output, input-collection, or condition component. These cover common tasks so you avoid writing custom code for routine steps.

A reusable code-backed building block that implements logic the built-in components cannot handle, such as calling a REST API or transforming data. It is registered with the skill and invoked like any other component.

The configuration that connects a skill to external systems, such as a REST endpoint or a database, so the conversation can read and write real data. It defines the request, response mapping, and error handling.

A declarative integration that describes an external HTTP API, including endpoint, method, headers, and payload schema, so the skill can call it without hand-written code. Authentication is configured alongside it.

A registered external or internal endpoint that a skill can invoke, exposing operations the dialog can call. Services are defined once and reused across flows and skills.

A messaging surface where the assistant is available, such as a web widget, a messaging app, or a voice interface. Each channel has its own configuration, capabilities, and payload format.

Structured message content tailored to a channel's rendering abilities, such as cards or quick replies. Payloads must degrade gracefully on channels that do not support the richer format.

An HTTP callback endpoint that a channel or external system calls to deliver events or messages to the assistant. Webhooks must be secured and validated to prevent spoofed traffic.

The mechanism proving identity for users or backend calls, such as OAuth, API keys, or channel-specific verification. It is configured per channel and per integration and must never be hard-coded in flows.

The rules that determine what an authenticated user or service may access, such as role-based permissions on skills or data. It complements authentication, which only establishes who the caller is.

A named set of permissions assigned to users or service accounts, controlling who can build, publish, or administer assistants. Least-privilege role design limits the blast radius of a compromised account.

The overall practice of protecting the assistant, its data, and its integrations, covering transport encryption, secret management, input validation, and audit logging. It spans design, build, and runtime operations.

Storing credentials such as API keys and passwords in a protected vault rather than in flow definitions or source code. Secrets are referenced at runtime so they can be rotated without editing the assistant.

The process of promoting a built assistant or skill from a development environment to test and production, including publishing versions and attaching channels. It should be repeatable and reversible.

An immutable snapshot of a skill or assistant at a point in time, used to publish and roll back safely. Channels point at a specific version, so changes stay isolated until explicitly promoted.

A separate instance of the platform, such as development, test, or production, with its own configuration and data. Promotion between environments moves artifacts, not live state.

The step that processes utterances and entity examples so the model can classify intents and extract values. Retraining is required after meaningful changes and should be validated with test phrases.

The response sent when the assistant cannot understand or handle the input, guiding the user toward a supported request. Well-written fallbacks reduce dead ends and abandonment.

Handing the conversation to a human agent or another system when the assistant cannot resolve the request. It requires passing context so the human does not restart the conversation from scratch.

The bounded interaction between one user and the assistant, holding conversation state until it ends or times out. Session scope determines which variables and context survive between turns.