JSPM

  • Created
  • Published
  • Downloads 4270
  • Score
    100M100P100Q128904F
  • License MIT

A TypeScript client library for connection to the Automattic agent framework.

Package Exports

  • @automattic/agenttic-client
  • @automattic/agenttic-client/mocks

Readme

@automattic/agenttic-client

A TypeScript client library for connecting to the Automattic agent framework with React hooks and component integration.

Original based on the A2A (Agent2Agent), but it has morphed to meet our internal needs so there is no longer a direct relationship to that protocol.

Installation

npm install @automattic/agenttic-client

Key Features

  • React hooks for agent communication (useAgentChat, useClientContext, useClientTools, useClientAbilities)
  • Streaming and non-streaming responses
  • Tool execution system with automatic handling
  • Compatibility with WordPress Abilities API
  • Context injection for each message
  • Conversation persistence and management
  • Request cancellation with abort controllers
  • TypeScript support
  • Message actions and markdown rendering

Quick Start

Basic Chat Integration

import { useAgentChat } from '@automattic/agenttic-client';

function ChatComponent() {
  const {
    messages,
    isProcessing,
    error,
    onSubmit,
    registerSuggestions,
    registerMarkdownComponents
  } = useAgentChat({
    agentId: 'big-sky',
    sessionId: 'my-session-123',
    authProvider: async () => ({ Authorization: 'Bearer your-token' })
  });

  return (
    <div>
      {messages.map(msg => (
        <div key={msg.id} className={msg.role}>
          {msg.content.map(content => content.text).join('')}
        </div>
      ))}
      <input
        onKeyDown={(e) => e.key === 'Enter' && onSubmit(e.target.value)}
        disabled={isProcessing}
      />
      {error && <div>Error: {error}</div>}
    </div>
  );
}

With Tools and Context

import { useAgentChat, useClientTools, useClientContext } from '@automattic/agenttic-client';

function AdvancedChatComponent() {
  const contextProvider = useClientContext(() => ({
    page: window.location.href,
    timestamp: Date.now(),
    userRole: getCurrentUserRole()
  }));

  const toolProvider = useClientTools(
    async () => [
      {
        id: 'calculator',
        name: 'Calculator',
        description: 'Perform mathematical calculations',
        input_schema: {
          type: 'object',
          properties: {
            expression: { type: 'string' }
          },
          required: ['expression']
        }
      }
    ],
    async (toolId, args) => {
      if (toolId === 'calculator') {
        return { result: calc(args.expression)};
      }
    }
  );

  const chat = useAgentChat({
    agentId: 'big-sky',
    contextProvider,
    toolProvider,
    authProvider: async () => ({ Authorization: 'Bearer token' })
  });

  return <ChatInterface {...chat} />;
}

With Abilities

WordPress Abilities API integration allows you to expose WordPress capabilities as tools to agents.

import { useAgentChat, useClientAbilities } from '@automattic/agenttic-client';
import { getAbilities, executeAbility } from '@wordpress/abilities';

function WordPressChat() {
  const [abilities, setAbilities] = useState([]);

  useEffect(() => {
    getAbilities().then(setAbilities);
  }, []);

  const toolProvider = useClientAbilities(abilities, executeAbility);

  const chat = useAgentChat({
    agentId: 'wp-assistant',
    toolProvider,
    authProvider: async () => ({ Authorization: 'Bearer token' })
  });

  return <ChatInterface {...chat} />;
}

You can also combine regular tools with abilities using useClientToolsWithAbilities:

const toolProvider = useClientToolsWithAbilities( {
    getClientTools: async () => myCustomTools,
    executeTool: async ( toolId, args ) => {
        /* execute custom tools */
    },
    abilities,
    executeAbility,
} );

Core APIs

useAgentChat Hook

The primary hook for chat functionality, providing everything needed for a complete chat interface.

const {
    // Chat state
    messages, // UIMessage[] - formatted for display
    isProcessing, // boolean - request in progress
    error, // string | null - last error

    // Core methods
    onSubmit, // (message: string) => Promise<void>
    abortCurrentRequest, // () => void - cancel in-flight request

    // Configuration
    registerSuggestions, // (suggestions: Suggestion[]) => void
    registerMarkdownComponents, // (components: MarkdownComponents) => void
    registerMessageActions, // (registration: MessageActionsRegistration) => void

    // Utilities
    messageRenderer, // React component for markdown rendering
    addMessage, // (message: UIMessage) => void
} = useAgentChat( config );

Config Options:

  • agentId: string - Required. Agent identifier
  • agentUrl?: string - Agent endpoint URL (defaults to WordPress.com)
  • sessionId?: string - Session ID for conversation persistence
  • contextProvider?: ContextProvider - Dynamic context injection
  • toolProvider?: ToolProvider - Tool execution capabilities
  • authProvider?: AuthProvider - Authentication headers

useClientContext Hook

Provides dynamic context that refreshes with each message.

const contextProvider = useClientContext( () => ( {
    currentPage: {
        url: window.location.href,
        title: document.title,
        selectedText: getSelection(),
    },
    user: {
        role: getUserRole(),
        permissions: getPermissions(),
    },
    timestamp: Date.now(),
} ) );

useClientTools Hook

Enables agents to execute tools in your application.

const toolProvider = useClientTools(
    // Define available tools
    async () => [
        {
            id: 'file-reader',
            name: 'File Reader',
            description: 'Read file contents',
            input_schema: {
                type: 'object',
                properties: {
                    path: { type: 'string' },
                },
                required: [ 'path' ],
            },
        },
    ],
    // Execute tool calls
    async ( toolId, args ) => {
        if ( toolId === 'file-reader' ) {
            return { content: await readFile( args.path ) };
        }
        throw new Error( `Unknown tool: ${ toolId }` );
    }
);

useClientAbilities Hook

Converts WordPress Abilities to Agenttic tools automatically.

import { getAbilities, executeAbility } from '@wordpress/abilities';

const abilities = await getAbilities();
const toolProvider = useClientAbilities( abilities, executeAbility );

WordPress Abilities can be:

  • Server-side: Executed via REST API (no callback property)
  • Client-side: Executed in browser (has callback function)

The API handles both types automatically, routing execution appropriately.

useClientToolsWithAbilities Hook

Combines regular tools and WordPress Abilities in a single provider.

import { getAbilities, executeAbility } from '@wordpress/abilities';

const abilities = await getAbilities();
const toolProvider = useClientToolsWithAbilities( {
    getClientTools: async () => [
        // Your regular tools
        { id: 'calculator', name: 'Calculator' /* ... */ },
    ],
    executeTool: async ( toolId, args ) => {
        // Handle regular tool execution
    },
    abilities,
    executeAbility,
} );

createClient Function

Low-level client for direct communication without React.

import { createClient } from '@automattic/agenttic-client';

const client = createClient({
  agentId: 'big-sky',
  authProvider: async () => ({ Authorization: 'Bearer token' }),
  toolProvider: {
    getAvailableTools: async () => [...],
    executeTool: async (toolId, args) => ({ result: '...' })
  }
});

// Non-streaming
const task = await client.sendMessage({
  message: createTextMessage('Hello'),
  sessionId: 'session-123'
});

// Streaming
for await (const update of client.sendMessageStream({
  message: createTextMessage('Hello'),
  sessionId: 'session-123'
})) {
  console.log(update.text);
  if (update.final) break;
}

// With abort control
const abortController = new AbortController();
const task = await client.sendMessage({
  message: createTextMessage('Hello'),
  sessionId: 'session-123',
  abortSignal: abortController.signal
});

// Cancel the request
abortController.abort();

Agent Manager

Functional singleton for managing multiple agent instances.

import { getAgentManager } from '@automattic/agenttic-client';

const manager = getAgentManager();

// Create agent
await manager.createAgent( 'my-agent', {
    agentId: 'big-sky',
    sessionId: 'session-123',
    contextProvider,
    toolProvider,
} );

// Send messages
const task = await manager.sendMessage( 'my-agent', 'Hello' );

// Streaming
for await ( const update of manager.sendMessageStream( 'my-agent', 'Hello' ) ) {
    console.log( update );
}

// Manage conversation
const history = manager.getConversationHistory( 'my-agent' );
await manager.resetConversation( 'my-agent' );

Additional Features

Request Cancellation

Cancel in-flight requests using the built-in abort functionality:

const { abortCurrentRequest, isProcessing } = useAgentChat( config );

// Cancel current request
if ( isProcessing ) {
    abortCurrentRequest();
}

For low-level client usage, use AbortController:

import {
    createClient,
    createAbortController,
} from '@automattic/agenttic-client';

const client = createClient( config );
const abortController = createAbortController();

// Start a request with abort signal
const requestPromise = client.sendMessage( {
    message: createTextMessage( 'Hello' ),
    abortSignal: abortController.signal,
} );

// Cancel the request
abortController.abort();

// Handle cancellation
try {
    await requestPromise;
} catch ( error ) {
    if ( error.name === 'AbortError' ) {
        console.log( 'Request was cancelled' );
    }
}

Message Actions

Add interactive buttons to agent messages:

const { registerMessageActions, createFeedbackActions } =
    useAgentChat( config );

// Built-in feedback actions
registerMessageActions(
    createFeedbackActions( {
        onFeedback: async ( messageId, feedback ) => {
            console.log( `${ feedback } feedback for ${ messageId }` );
        },
        icons: { up: '👍', down: '👎' },
    } )
);

// Custom actions
registerMessageActions( {
    id: 'copy-actions',
    actions: [
        {
            id: 'copy',
            label: 'Copy',
            icon: '📋',
            onClick: ( message ) =>
                navigator.clipboard.writeText( message.content[ 0 ].text ),
        },
    ],
} );

Markdown Extensions

Extend markdown rendering with custom components:

import { BarChart, LineChart } from '@automattic/agenttic-client';

const { registerMarkdownComponents, registerMarkdownExtensions } = useAgentChat(config);
// Register chart components
registerMarkdownComponents( {
  // Custom heading styles
  h1: ({ children }) => (
    <h1 className="text-2xl font-bold text-brand">{children}</h1>
  ),
  // Custom code blocks with syntax highlighting
  code: ({ children, className }) => (
    <SyntaxHighlighter language={className}>
    {children}
    </SyntaxHighlighter>
  ),
  // Custom link handling
  a: ({ href, children }) => (
    <a href={href} target="_blank" rel="noopener">
    {children}</a>
  ),
  blockquote: ( { children, ...props } ) => (
    <blockquote
      { ...props }
      style={ {
        borderLeft: '4px solid #007cba',
        backgroundColor: '#f0f8ff',
        margin: '16px 0',
        padding: '12px 16px',
        fontStyle: 'italic',
        borderRadius: '0 4px 4px 0',
      } }
    >
      { children }
    </blockquote>
  ),
});

// Register custom extensions
registerMarkdownExtensions({
  charts: {
    BarChart,
    LineChart
  }
});

Conversation Suggestions

Provide suggested prompts to users:

const { registerSuggestions, clearSuggestions } = useAgentChat( config );

registerSuggestions( [
    {
        id: '1',
        label: 'Help me write code',
        prompt: 'Can you help me write a function?',
    },
    {
        id: '2',
        label: 'Explain this error',
        prompt: 'What does this error mean?',
    },
] );

Type Definitions

interface UIMessage {
    id: string;
    role: 'user' | 'agent';
    content: Array< {
        type: 'text' | 'image_url' | 'component';
        text?: string;
        image_url?: string;
        component?: React.ComponentType;
    } >;
    timestamp: number;
    actions?: UIMessageAction[];
}

interface Tool {
    id: string;
    name: string;
    description: string;
    input_schema: {
        type: 'object';
        properties: Record< string, any >;
        required?: string[];
    };
}

type AuthProvider = () => Promise< Record< string, string > >;
type ContextProvider = { getClientContext: () => any };
type ToolProvider = {
    getAvailableTools: () => Promise< Tool[] >;
    executeTool: ( toolId: string, args: any ) => Promise< any >;
};

Development

# Build the package
pnpm build

# Run tests
pnpm test

# Type checking
pnpm type-check

# Lint
pnpm lint