@inferable/assistant-ui

Assistant UI Runtime

Inferable Runtime for assistant-ui.

Installation

Install the package and its peer dependencies:

npm install @inferable/assistant-ui @assistant-ui/react

yarn add @inferable/assistant-ui @assistant-ui/react

pnpm add @inferable/assistant-ui @assistant-ui/react

Quick Start

More details on Inferable front-end usage can be found here.

useInferableRuntime

useInferableRuntime provides an AssistantRuntime object which can be used with assistant-ui to render a Chat UI with Inferable.

import { useInferableRuntime } from '@inferable/assistant-ui';
import { AssistantRuntimeProvider, Thread } from "@assistant-ui/react";

type RuntimeOptions = {
  clusterId: string;
  baseUrl?: string;
  runId?: string;
} & (
  | { apiSecret: string; customAuthToken?: never }
  | { customAuthToken: string; apiSecret?: never }
);

const { runtime, run } = useInferableRuntime({
  clusterId: '<YOUR_CLUSTER_ID>',
  // Choose one of the following authentication methods:

  // Option 1: Custom Auth Token (Recommended)
  customAuthToken: 'your-custom-auth-token',

  // Option 2: API Secret (Not recommended for browser usage)
  // apiSecret: 'your-api-secret',

  // Optional configuration
  baseUrl: 'https://api.inferable.ai', // Optional: defaults to production URL
  runId: 'existing-run-id', // Optional: to resume an existing run
})

return (
  <div className="h-full">
    <AssistantRuntimeProvider runtime={runtime}>
      <Thread/>
    </AssistantRuntimeProvider>
  </div>
);

userInferableRuntime can also be used with AssistantModal for a modal UI.

Configuration Options

The useInferableRuntime hook accepts the following options:

• clusterId (required): The ID of your Inferable cluster
• Authentication (one of the following is required):
  • customAuthToken: A custom authentication token (recommended for browser usage)
  • apiSecret: Your cluster's API secret (not recommended for browser usage)
• Optional configuration:
  • baseUrl: The base URL for API requests (defaults to production URL)
  • runId: An existing run ID to resume

Rendering function UI

You can provide assistant-ui with custom UI components for rendering Inferable function calls / results.

Example

// Fallback UI
const FallbackToolUI = ({args, result, toolName}) =>
  <div className="center">
    <h1>Tool: {toolName}</h1>
    <h2>Input:</h2>
    <pre className="whitespace-pre-wrap">{JSON.stringify(args, null, 2)}</pre>
    <h2>Output:</h2>
    {result && <pre className="whitespace-pre-wrap">{JSON.stringify(result, null, 2)}</pre>}
    {!result && <p>No output</p>}
  </div>

// Custom UI example
const SearchToolUI = makeAssistantToolUI<any, any>({
  toolName: "default_webSearch",
  render: ({ args }) => {
    return <p>webSearch({args.query})</p>;
  },
});

return (
  <div className="h-full">
    <AssistantRuntimeProvider runtime={runtime}>
      <Thread
        tools={[
          WebSearchToolUI
        ]},
        assistantMessage={{
          components: {
            ToolFallback: FallbackToolUI
          },
      }} />
    </AssistantRuntimeProvider>
  </div>
);

Local Development

There is development server included in the repository at ./demo.

1. Start the development server:

npm run dev

This will start a Vite dev server at http://localhost:3000 with the test page, which provides a simple interface to test the runtime.

Documentation

• Inferable documentation contains all the information you need to get started with Inferable.
• Assistant UI documentation contains all the information you need to get started with Assistant UI.

Support

For support or questions, please create an issue in the repository.

Contributing

Contributions to the Inferable React SDK are welcome. Please ensure that your code adheres to the existing style and includes appropriate tests.