react-native-ai-toolkit

AI chat toolkit for React Native — SSE streaming, chat hooks, UI components, and parsers for any LLM backend.

Why

React Native doesn't support the browser EventSource API, and there's no comprehensive solution for building AI chat features with streaming responses. react-native-ai-toolkit provides everything in one package:

• SSE client with automatic reconnection
• React hooks for streaming and chat management
• Ready-made UI components
• Parsers for structured data extraction from AI responses

Works with any LLM backend — OpenAI, Anthropic, custom APIs. No framework lock-in.

Requirements

• React Native >= 0.71
• React >= 18

Installation

npm install react-native-ai-toolkit

Quick Start

Never hardcode API keys. Use environment variables or a proxy backend.

import { useChat, ChatList, ChatBubble } from 'react-native-ai-toolkit';

function ChatScreen() {
  const { messages, sendMessage, isStreaming, tokenUsage } = useChat({
    apiUrl: 'https://api.openai.com/v1/chat/completions',
    systemPrompt: 'You are a helpful assistant.',
    headers: {
      'Content-Type': 'application/json',
      Authorization: `Bearer ${process.env.EXPO_PUBLIC_OPENAI_KEY}`,
    },
  });

  return (
    <ChatList
      messages={messages}
      renderMessage={(msg) => (
        <ChatBubble
          message={msg}
          variant={msg.role === 'user' ? 'user' : 'assistant'}
        />
      )}
    />
  );
}

Hooks

useChat

High-level hook for chat management. Handles messages, history, streaming, retry, and token tracking.

const {
  messages,      // Message[]
  sendMessage,   // (content: string) => void
  isStreaming,   // boolean
  tokenUsage,    // { promptTokens, completionTokens, totalTokens } | null
  retry,         // () => void
  clear,         // () => void
  error,         // Error | null
  stop,          // () => void
} = useChat(config);

Config:

| Property | Type | Default | Description | |----------|------|---------|-------------| | apiUrl | string | — | Chat API endpoint | | systemPrompt | string | — | System message | | headers | Record<string, string> | — | Request headers | | model | string | — | LLM model name | | initialMessages | Message[] | — | Pre-loaded messages | | buildRequestBody | (messages) => object | OpenAI format | Custom request builder | | parseResponse | (chunk) => string | OpenAI format | Custom response parser |

useAIStream

Low-level hook for raw SSE streaming. Use when you need more control than useChat provides.

const {
  text,          // string — accumulated response text
  status,        // 'idle' | 'connecting' | 'streaming' | 'done' | 'error'
  error,         // Error | null
  tokenUsage,    // TokenUsage | null
  send,          // (body?) => void
  abort,         // () => void
  reset,         // () => void
} = useAIStream(config);

Cleanup & Cancellation

useChat automatically aborts the SSE connection on component unmount. Call stop() to cancel a streaming response manually:

<Button onPress={stop} title="Cancel" />

Components

ChatList

Message list with auto-scroll to bottom.

<ChatList
  messages={messages}
  renderMessage={(msg) => <ChatBubble message={msg} variant="assistant" />}
  renderStreamingIndicator={() => <ActivityIndicator />}
  style={{ flex: 1 }}
  flatListProps={{ inverted: false }}
/>

ChatBubble

Message bubble with user/assistant variants.

<ChatBubble
  message={message}
  variant="user"
  renderContent={(text) => <Markdown>{text}</Markdown>}
  showAvatar
/>

StreamingText

Text that "types" as tokens arrive. Markdown rendering is deferred until the stream completes to avoid layout jumps.

<StreamingText
  text={streamingText}
  showCursor
  renderContent={(text) => <Markdown>{text}</Markdown>}
/>

Parsers

Pure functions for extracting structured data from AI responses. Use outside of React — in utilities, middleware, or custom hooks.

parseContent(data)

Extracts text from OpenAI-style streaming delta.

parseContent({ choices: [{ delta: { content: 'Hello' } }] })
// => 'Hello'

extractTokenUsage(data)

Extracts token counts from the response.

extractTokenUsage({ usage: { prompt_tokens: 24, completion_tokens: 20, total_tokens: 44 } })
// => { promptTokens: 24, completionTokens: 20, totalTokens: 44 }

extractJSON(text)

Extracts JSON from AI response text (fenced code blocks or raw JSON).

extractJSON('Result: ```json\n{"score": 0.95}\n```')
// => { score: 0.95 }

extractToolCalls(data)

Extracts function calling data from OpenAI tool_calls or raw JSON arrays.

extractToolCalls({ tool_calls: [{ type: 'function', function: { name: 'search', arguments: '{"q":"test"}' } }] })
// => [{ type: 'function', name: 'search', input: { q: 'test' } }]

extractReasoning(text)

Separates chain-of-thought reasoning from the main content. Supports <think />, <reasoning />, and <thought /> tags.

// DeepSeek-style format
extractReasoning('<think source="internal">Thinking...</think >Answer: 42.')
// => { reasoning: 'Thinking...', content: 'Answer: 42.' }

Custom Backends

buildRequestBody and parseResponse let you adapt the hook to any API format:

const { messages, sendMessage } = useChat({
  apiUrl: 'https://my-api.com/chat',
  buildRequestBody: (msgs) => ({
    messages: msgs,
    model: 'my-model',
    stream: true,
  }),
  parseResponse: (chunk) => {
    const parsed = JSON.parse(chunk);
    return parsed.result;
  },
  headers: { Authorization: 'Bearer TOKEN' },
});

Using SSE Directly

For maximum control, use setupEventSource directly. The onMessage callback returns an object with optional tokens — this value is used internally to track token usage:

import { setupEventSource } from 'react-native-ai-toolkit';

const es = setupEventSource(
  { url: '/api/stream', headers: { Authorization: 'Bearer TOKEN' } },
  {
    onOpen: () => console.log('Connected'),
    onMessage: (data, parsed) => {
      // Return { tokens } to accumulate usage, or nothing to skip tracking
      return { tokens: parsed.usage?.total_tokens };
    },
    onComplete: (totalTokens) => console.log('Done, tokens:', totalTokens),
    onError: (error) => console.error(error),
  }
);

// Close when done
es.close();

Contributing

• Development workflow
• Sending a pull request
• Code of conduct

License

MIT