useChat

useChat(options?: object): UseChatResult

A React hook for managing chat completions with authentication.

React Native version - This is a lightweight version that only supports API-based chat completions. Local chat and client-side tools are not available in React Native.

Parameters

Parameter	Type	Description
`options?`	`object`	Optional configuration object
`options.apiType?`	`ApiType`	Which API endpoint to use. Default: “responses” “responses”: OpenAI Responses API (supports thinking, reasoning, conversations) “completions”: OpenAI Chat Completions API (wider model compatibility)
`options.baseUrl?`	`string`	Optional base URL for the API requests.
`options.getToken?`	() => `Promise`<`string` \| `null`>	An async function that returns an authentication token.
`options.onData?`	(`chunk`: `string`) => `void`	Callback function to be called when a new data chunk is received.
`options.onError?`	(`error`: `Error`) => `void`	Callback function to be called when an unexpected error is encountered. Note: This callback is NOT called for aborted requests (via `stop()` or component unmount). Aborts are intentional actions and are not considered errors. To detect aborts, check the `error` field in the `sendMessage` result: `result.error === "Request aborted"`.
`options.onFinish?`	(`response`: `ApiResponse`) => `void`	Callback function to be called when the chat completion finishes successfully. Receives raw API response - either Responses API or Completions API format.
`options.onServerToolCall?`	(`toolCall`: `ServerToolCallEvent`) => `void`	Callback function to be called when a server-side tool (MCP) is invoked during streaming. Use this to show activity indicators like “Searching…” in the UI.
`options.onThinking?`	(`chunk`: `string`) => `void`	Callback function to be called when thinking/reasoning content is received. This is called with delta chunks as the model “thinks” through a problem.
`options.onToolCall?`	(`toolCall`: `LlmapiToolCall`) => `void`	Callback function to be called when a tool call is requested by the LLM. This is called for tools that don’t have an executor or have autoExecute=false. The app should execute the tool and send the result back.
`options.smoothing?`	`boolean` \| `StreamSmoothingConfig`	Controls adaptive output smoothing for streaming responses. Fast models can return text faster than is comfortable to read — smoothing buffers incoming chunks and releases them at a consistent, adaptive pace. `true` or omitted: enabled with defaults (200→400 chars/sec over 3s) `false`: disabled, callbacks fire immediately with raw chunks `StreamSmoothingConfig`: custom speed/ramp configuration Default `true`

Returns

UseChatResult

An object containing:

isLoading: A boolean indicating whether a request is currently in progress
sendMessage: An async function to send chat messages
stop: A function to abort the current request

Example


const { isLoading, sendMessage, stop } = useChat({
  getToken: async () => await getAuthToken(),
  onFinish: (response) => console.log("Chat finished:", response),
  onError: (error) => console.error("Chat error:", error)
});
 
const handleSend = async () => {
  const result = await sendMessage({
    messages: [{ role: 'user', content: [{ type: 'text', text: 'Hello!' }] }],
    model: 'gpt-4o-mini'
  });
};