@redhat-cloud-services/ai-client-common

@redhat-cloud-services/ai-client-common

Common interfaces and utilities for AI client packages in the Red Hat Cloud Services ecosystem.

Features

• Standardized AI Client Interface - Common IAIClient interface for all AI services
• Conversation Management - Standard conversation interface with locking support
• Dependency Injection - Interfaces for custom fetch implementations and streaming handlers
• TypeScript Support - Comprehensive type definitions for AI client development
• Error Handling - Base error classes with validation error support
• Streaming Support - Generic streaming handler interface for real-time responses
• Zero Dependencies - Pure TypeScript interfaces with no runtime dependencies

Installation

npm install @redhat-cloud-services/ai-client-common

Core Interfaces

IAIClient Interface

All AI clients in this workspace implement the IAIClient interface:

import { IAIClient, ClientInitLimitation, IInitErrorResponse } from '@redhat-cloud-services/ai-client-common';

declare class IAIClient<AP extends Record<string, unknown> = Record<string, unknown>, TChunk = unknown> {
  constructor(config: IBaseClientConfig<TChunk>);
  
  init(): Promise<{
    conversations: IConversation[];
    limitation?: ClientInitLimitation;
    error?: IInitErrorResponse;
  }>;
  
  // Basic message sending
  sendMessage<T extends Record<string, unknown> = Record<string, unknown>>(
    conversationId: string, 
    message: string, 
    options?: ISendMessageOptions<T>
  ): Promise<IMessageResponse<AP>>;
  
  // Message sending with custom request payload
  sendMessage<
    T extends Record<string, unknown> = Record<string, unknown>,
    R extends Record<string, unknown> = Record<string, unknown>
  >(
    conversationId: string, 
    message: string, 
    options?: ISendMessageOptions<T, R>
  ): Promise<IMessageResponse<AP>>;
  
  getDefaultStreamingHandler<TChunk = unknown>(): IStreamingHandler<TChunk> | undefined;
  
  getConversationHistory(conversationId: string, options?: IRequestOptions): Promise<IConversationHistoryResponse<AP>>;
  
  healthCheck(options?: IRequestOptions): Promise<unknown>;
  
  getServiceStatus?(options?: IRequestOptions): Promise<unknown>;
  
  createNewConversation(): Promise<IConversation>;
}

Dependency Injection

Custom Fetch Implementation

Important: Do NOT set 'Content-Type' headers in your fetchFunction - AI clients manage these internally based on endpoint requirements.

import { IFetchFunction } from '@redhat-cloud-services/ai-client-common';

const customFetch: IFetchFunction = async (input, init) => {
  // Add authentication headers
  const token = await getAuthToken();
  
  return fetch(input, {
    ...init,
    headers: {
      ...init?.headers,
      'Authorization': `Bearer ${token}`,
      // DO NOT set 'Content-Type' - AI clients handle this internally
    },
  });
};

Base Client Configuration

import { IBaseClientConfig } from '@redhat-cloud-services/ai-client-common';

const config: IBaseClientConfig = {
  baseUrl: 'https://your-ai-service.com',
  fetchFunction: customFetch, // Optional - defaults to native fetch
  defaultStreamingHandler: new CustomStreamingHandler() // Optional
};

Lazy Initialization (Default Behavior)

AI clients now use lazy initialization by default. The init() method no longer auto-creates conversations - instead, conversations are created automatically when needed (e.g., on first message send).

Default Behavior:

• Client init() only loads existing conversations
• No conversations are auto-created during initialization
• Conversations are created automatically on first sendMessage() call
• Provides optimal performance and seamless user experience

Key Changes:

• Removed: ClientInitOptions interface (no longer needed)
• Removed: getInitOptions() method from IAIClient interface
• Removed: initialConversationId from init() return type
• New: Automatic conversation creation on first message send

Client Initialization Responses

The init() method returns additional information about client limitations and errors:

import { ClientInitLimitation, IInitErrorResponse } from '@redhat-cloud-services/ai-client-common';

// Client limitations (e.g., quota exceeded)
type ClientInitLimitation = {
  reason: string;
  detail?: string;
};

// Initialization errors
interface IInitErrorResponse {
  message: string;
  status: number;
}

// Utility function to check if an object is an IInitErrorResponse
function isInitErrorResponse(obj: unknown): obj is IInitErrorResponse;

Streaming Support

The streaming interface has been updated to standardize chunk handling across all AI clients. The afterChunk callback now receives an IStreamChunk object with standardized structure.

IStreamChunk Interface

import { IStreamChunk } from '@redhat-cloud-services/ai-client-common';

interface IStreamChunk<T extends Record<string, unknown> = Record<string, unknown>> {
  answer: string;
  messageId: string;
  additionalAttributes: T;
}

Implementing a Custom Streaming Handler

import { IStreamingHandler, AfterChunkCallback, IStreamChunk } from '@redhat-cloud-services/ai-client-common';

class CustomStreamingHandler<TChunk = unknown> implements IStreamingHandler<TChunk> {
  onChunk(chunk: TChunk, afterChunk?: AfterChunkCallback): void {
    console.log('Received chunk:', chunk);
    
    // Process the chunk and call afterChunk with standardized format
    if (afterChunk) {
      afterChunk({
        answer: extractAnswer(chunk), // Extract answer from chunk
        additionalAttributes: extractAttributes(chunk) // Extract additional data
      });
    }
  }

  onStart?(conversationId?: string, messageId?: string): void {
    console.log('Stream started', { conversationId, messageId });
  }

  onComplete?(finalChunk: TChunk): void {
    console.log('Stream completed:', finalChunk);
  }

  onError?(error: Error): void {
    console.error('Stream error:', error);
  }

  onAbort?(): void {
    console.log('Stream aborted');
  }
}

Send Message Options

import { ISendMessageOptions, IStreamChunk } from '@redhat-cloud-services/ai-client-common';

// Basic streaming options
const streamingOptions: ISendMessageOptions = {
  stream: true,
  headers: { 'Custom-Header': 'value' },
  signal: abortController.signal,
  afterChunk: (chunk: IStreamChunk) => {
    // Process each standardized chunk as it arrives
    console.log('Answer:', chunk.answer);
    console.log('Additional data:', chunk.additionalAttributes);
    updateUI(chunk.answer);
  }
};

// Options with custom request payload (client-specific)
const optionsWithPayload: ISendMessageOptions<AdditionalProps, CustomPayload> = {
  stream: false,
  headers: { 'X-Custom': 'value' },
  requestPayload: {
    // Client-specific payload data
    customData: 'value',
    options: { setting: true }
  }
};

Request Payload Support

The ISendMessageOptions interface supports method overloading to enable client-specific request payloads:

// Interface definition
export interface ISendMessageOptions<
  T extends Record<string, unknown> = Record<string, unknown>,
  R extends Record<string, unknown> = never
> extends IRequestOptions {
  stream?: boolean;
  afterChunk?: AfterChunkCallback<T>;
  requestPayload?: R extends never ? never : R;
}

// Usage with client-specific payload
interface MyClientPayload {
  context?: { systemInfo: string };
  skipCache?: boolean;
}

const response = await client.sendMessage(
  'conversation-id',
  'message',
  {
    requestPayload: {
      context: { systemInfo: 'linux' },
      skipCache: true
    }
  }
);

Error Handling

Base Error Classes

import { 
  AIClientError, 
  AIClientValidationError 
} from '@redhat-cloud-services/ai-client-common';

try {
  const response = await client.sendMessage(conversationId, message);
} catch (error) {
  if (error instanceof AIClientValidationError) {
    console.error('Validation errors:', error.validationErrors);
    error.validationErrors.forEach(validationError => {
      console.log(`Field: ${validationError.loc.join('.')}`);
      console.log(`Message: ${validationError.msg}`);
      console.log(`Type: ${validationError.type}`);
    });
  } else if (error instanceof AIClientError) {
    console.error(`API Error ${error.status}: ${error.message}`);
    console.error('Response data:', error.data);
  } else {
    console.error('Unexpected error:', error);
  }
}

Type Definitions

Request and Response Types

import { 
  IRequestOptions,
  IMessageResponse,
  IConversationHistoryResponse,
  IConversation
} from '@redhat-cloud-services/ai-client-common';

// Standard request options
const options: IRequestOptions = {
  headers: { 'Custom-Header': 'value' },
  signal: new AbortController().signal
};

// Message response structure
interface IMessageResponse<AP = Record<string, unknown>> {
  messageId: string;
  answer: string;
  conversationId: string;
  date?: Date;
  additionalAttributes?: AP;
}

// Conversation structure
interface IConversation {
  id: string;
  title: string;
  locked: boolean; // Prevents new messages when true
  createdAt: Date;
}

Compatible Packages

This package provides the foundation for:

• @redhat-cloud-services/arh-client - Intelligent Front Door (IFD) API client
• @redhat-cloud-services/lightspeed-client - OpenShift Lightspeed API client
• @redhat-cloud-services/ansible-lightspeed - Ansible Lightspeed API client
• @redhat-cloud-services/rhel-lightspeed-client - RHEL LightSpeed RAG API client
• @redhat-cloud-services/aai-client - Ansible Assisted Installer API client
• @redhat-cloud-services/ai-client-state - State management for AI conversations
• @redhat-cloud-services/ai-react-state - React hooks and context for AI state

Building

Run nx build ai-client-common to build the library.

Running unit tests

Run nx test ai-client-common to execute the unit tests via Jest.

Development

This package follows the workspace standards:

• Strict TypeScript with no any types
• Zero runtime dependencies
• Comprehensive interface definitions