Component

The component definition describes a reusable integration component (its metadata, actions, triggers, and optional connection). Below is an explanation of each method you can use when building a component with the DSL:

• component(String name) - Builds a new ModifiableComponentDefinition with the specified backend ID. The name is the component ID used internally and in workflow JSON.
• actions(A... actionDefinitions) - Declares the actions exposed by this component.
• agentChannels(ModifiableAgentChannelDefinition... agentChannels) - Declares that this component can carry an AI agent's conversation. See Agent channels below.
• categories(ComponentCategory... category) / categories(List<ComponentCategory> categories) - Assigns UI categories for grouping and discovery in the catalog. See ComponentCategory.
• connection(ModifiableConnectionDefinition connectionDefinition) - Defines the connection used by actions/triggers. If not set, actions can still run without a connection if they don't require one.
• customAction(boolean customAction) - If true, enables the Custom Action feature in the UI (typically for REST/OpenAPI‑based connectors).
• customActionHelp(Help customActionHelp) - Adds help text for the Custom Action. Displayed as a popup in the UI next to the Custom Action.
• description(String description) - Short description shown in the catalog and tooltips.
• icon(String icon) - Path or resource name of the SVG icon shown in the UI.
• resources(String documentationUrl) / resources(String documentationUrl, Map<String, String> additionalUrls) - Links to product documentation and optional additional resources (e.g., FAQ, blog posts, templates).
• title(String title) - Human‑readable display name (Chicago style) shown in the UI.
• triggers(T... triggerDefinitions) - Declares triggers available on this component.
• clusterElements(ClusterElementDefinition<?>... clusterElements) - Declares reusable cluster elements (e.g., AI tools) that the component may expose.
• version(int version) - Component version number. Increment this when you make breaking changes so existing workflows can continue to use older versions.

Example

private static final ComponentDefinition COMPONENT_DEFINITION = component("textHelper")
        .title("Text Helper")
        .description("Helper component which contains operations to help you work with text.")
        .icon("path:assets/text-helper.svg")
        .categories(ComponentCategory.HELPERS)
        .actions(TextHelperUpperCaseAction.ACTION_DEFINITION);

Agent channels

An agent channel is a way an AI agent can be reached and can answer: a trigger that receives an incoming message and, usually, an action that sends the reply. Declaring one makes your component selectable as a channel on any agent - no change is needed anywhere else in the platform.

The channel contract is three fields. Each end of the channel states where they live on it, in its own vocabulary:

Because the request side defaults to the contract's own names, a trigger whose output already is the contract (build it with ComponentDsl.agentChannelRequest()) needs nothing but a bare agentRequest(). A trigger with its own payload shape spells the paths out, dots included - e.g. .conversationId("message.chat.id").

Pair them on the component with agentChannel(name, trigger[, replyAction]). The name is the channel's stored key and must be unique across all components:

// trigger
public static final ModifiableTriggerDefinition TRIGGER_DEFINITION = trigger("newChatRequest")
        .title("New Chat Request")
        .output(...)
        .agentRequest(agentRequest().attachments(ATTACHMENTS))
        .webhookRequest(ChatNewRequestTrigger::getWebhookResult);

// reply action
public static final ModifiableActionDefinition ACTION_DEFINITION = action("responseToRequest")
        .title("Response to Request")
        .agentReply(agentReply().attachments(ATTACHMENTS))
        .perform(ChatResponseToRequestAction::perform);

// component
private static final ComponentDefinition COMPONENT_DEFINITION = component("chat")
        .title("Chat")
        .agentChannels(
                agentChannel("chat", ChatNewRequestTrigger.TRIGGER_DEFINITION,
                        ChatResponseToRequestAction.ACTION_DEFINITION)
                                .title("Chat")
                                .approvalChannel("chat"));

Optional on agentChannel(...):

• title(String title) / description(String description) - How the channel is labelled in the agent UI. Defaults to the component's own title, which is wrong whenever one component exposes a channel that is not the whole component (Twilio's channel is titled "Twilio (WhatsApp)").
• approvalChannel(String elementName) - Which of this component's APPROVAL_CHANNELS cluster elements a human approval request is delivered through on this channel. Omit it and the channel simply cannot carry approvals.
• triggerParameters(Map<String, Object> triggerParameters) - Parameters pinned onto the generated trigger node, for a trigger whose output shape depends on its own input.

Optional on agentReply():

• channelParameter(String rowKey, String property) - A value the user configures on the channel itself is copied into property on every reply. Twilio's number becomes the reply's From this way; rowKey must name a property the paired trigger declares.
• fixedParameter(String property, Object value) - A reply parameter pinned by the declaration, e.g. useTemplate = false.

Everything is validated when the component loads: a path or property name that the paired trigger or action does not declare fails fast, rather than producing an agent that silently replies nowhere. A trigger that receives an event it should not act on declines by returning an empty collection from its webhook handler - the agent run never starts.