Actions

Actions are the bridge between Pillar's co-pilot and your application. They represent things users can do—like navigating to a page, opening a modal, or triggering a workflow—and the co-pilot suggests them when relevant.

Why Actions Matter

Without actions, the co-pilot can only answer questions. With actions, it becomes an assistant that helps users do things:

Without Actions	With Actions
"Here's how to invite a team member..."	"Here's how to invite a team member. [Invite Team Member]"
User reads instructions, navigates manually	User clicks the action, modal opens automatically

Actions turn help content into executable guidance.

How It Works

You define actions in your code

Describe what each action does in natural language the AI can understand.

User asks a question

"How do I change my billing plan?"

AI matches intent to actions

The co-pilot finds your "Upgrade Plan" action based on semantic similarity.

User clicks the suggested action

Your handler runs—navigates to billing, opens the upgrade modal, whatever you defined.

Action Types

Pillar supports five action types for different use cases:

Type	What It Does	Example
`navigate`	Goes to a page in your app	Settings, dashboard, detail pages
`trigger_action`	Runs your custom logic	Opens modals, starts wizards, toggles features
`inline_ui`	Shows interactive UI in the chat	Confirmation forms, data previews
`external_link`	Opens a URL in a new tab	Documentation, external resources
`copy_text`	Copies text to clipboard	API keys, code snippets

Auto-Run Behavior

Some actions run automatically without user confirmation:

navigate actions auto-run by default (just takes user to a page)
external_link actions auto-run (opens a new tab)
trigger_action requires user confirmation (executes custom logic)

You can override this with the autoRun property on any action.

Data Extraction

Actions can define a dataSchema that tells the co-pilot what data to extract from the conversation:

{
  name: 'invite_member',
  dataSchema: {
    email: { type: 'string', description: 'Email address to invite' },
    role: { type: 'string', enum: ['admin', 'member', 'viewer'] },
  },
}

When the action runs, your handler receives the extracted data:

onTask('invite_member', (data) => {
  // data.email and data.role are populated from the conversation
  inviteUser(data.email, data.role);
});

Context Requirements

Actions can specify required context to only appear in relevant situations:

An "Edit Project" action only appears when context.projectId is set
An "Admin Settings" action only appears when context.userRole === 'admin'

This keeps suggestions relevant and prevents users from seeing actions they can't execute.

Composability with Plans

When users ask for help with complex tasks, the co-pilot automatically chains your actions into Plans—multi-step checklists that guide users through workflows.

You don't need to define these workflows explicitly. Just define focused, single-purpose actions, and the AI composes them as needed.

Next Steps

Setting Up Actions — Step-by-step implementation guide
Action Types Reference — Detailed specs for each type
Plans — How actions combine into workflows

Search documentation