Back to DeploymentEnd users can interact with an AI Agent only after it is published. The platform provides a publishing workflow for administrators to review and approve changes before they reach end users.Before publishing, configure at least one channel. See Channel Enablement.
Default state when an agent is created. All changes happen here. Publishing a component creates a Published copy, but the In Development version remains editable.
Published
Contains components approved for end-user access. Developers cannot edit the Published copy directly.
Switch between states using the dropdown on the top left of the agent window.The Publish Status indicator tracks the progress of each publish request and reports success or failure with details.
If the agent is trained in a language using NLP Version 2, a banner notifies you about the upcoming auto-upgrade to NLP Version 3.
Configuration started but not complete. Applies to Alert, Action, and Information tasks only.
Configured
Configuration complete; not yet published.
Awaiting Approval
Publish request submitted; pending admin review.
Published
Approved and available to end users.
Upgrade in Progress
A new version of a published task is being configured.
Rejected
Admin rejected the publish request. A notification email is sent to the developer.
Suspended
Admin suspended the agent. End users cannot access it; developers can still work on the in-development copy.
A Published task cannot be deleted. Admins can Suspend it if it’s no longer needed. A Suspended task can be Upgraded to create a Configured copy for further changes.
Flows without a published version; upgraded versions of published flows
Tasks (Dialog, Alert, Panels, Widgets, Digital Forms, Small Talk)
Tasks without a published version; upgraded versions. When a new language is enabled in development, all tasks appear in the publish window with the new language checkbox selected.
Knowledge Graph (per language)
New and updated Level 1 terms; Traits and Synonyms replace the published set with all additions, updates, and deletions
NL ModelThe NL Model publishes training data for specific intents without publishing task definitions — letting NLP trainers and developers work independently.Steps to publish training data for specific intents:
Expand Natural Language > NL Model and click All Components.
Select the relevant components:
Component
Behavior
ML Utterances
Replaces old utterances with the In Development set. Click Customize to select utterances for specific tasks.
Patterns
Replaces old patterns. Click Customize to select updated patterns.
Rules
Replaces intents with rules for Dialog Execution and Knowledge Graph detection. Click Customize to select specific rules.
Trait Types
Replaces traits in the Published version with the In Development set.
Others
Replaces Bot Synonyms, Utterances and NER, Ignore Words and Field Memory, Entity Patterns and Negative Patterns, and Entity Synonyms.
If Dialog Tasks are selected under Publish > Tasks, their training data is published automatically and NL Model components cannot be customized separately. Deselect the tasks first to customize ML Utterances, Patterns, and Rules individually. Selecting NL Model components and then adding Dialog Tasks triggers a warning to reset publishable components.
Settings
Component
Behavior
Standard Responses
Replaces the full set with the In Development version
Default Dialog
Replaces the configuration with the In Development version
Amend Settings
Replaces the configuration with the In Development version
Advanced Settings
Replaces the configuration with the In Development version
The following components are outside the publishing workflow — they either don’t affect end users directly or only do so as part of a task definition. Both agent states show the same information for these:
Batch Testing (custom suites and test reports)
Manage Developer Access
App Creation, Resetting, and Deletion
Language enablement: Enabling a new language in the In Development version requires re-publishing all components for that language.
When you publish, the platform sends an approval request to the admin. Behavior depends on the agent’s Purpose:
Agent Type
Approval Outcome
Employee Agent
Admin selects users from the enterprise account who can access the agent, then approves. Only selected users can interact with the agent in the selected channels.
Consumer Agent
Any user can access the agent in the selected channels once the admin approves.
