Scraper Fox Gradient AI

A lightweight JavaScript client for interacting with chat.gradient.network using streaming responses.
Supports model selection, cluster mode configuration, and optional "thinking" reasoning mode.

Features

• Simple single-function interface: gradientAI(prompt, options)
• Supports multiple model presets
• Streams output and returns the final processed message
• Small, dependency-minimal, and ESM compatible
• Built on top of impit

Installation

pnpm add scraperfox-gradient-ai
# or
npm install scraperfox-gradient-ai
# or
yarn add scraperfox-gradient-ai

Usage

import { gradientAI } from "scraperfox-gradient-ai";

const result = await gradientAI("Hello AI!");
console.log(result);

With Custom Options

const response = await gradientAI("Explain quantum computing simply.", {
  model: "Qwen3 235B",
  enableThinking: true
});

console.log(response);

Available Configuration Options

| Option | Type | Default | Allowed Values | |-----------------|---------|-----------------|-----------------------------------------| | model | string | "GPT OSS 120B"| "GPT OSS 120B" | "Qwen3 235B" | | clusterMode | string | "hybrid" | "hybrid" | "nvidia" (NVIDIA GPT only) | | enableThinking| boolean | false | true or false |

Example

await gradientAI("Hello", {
  model: "GPT OSS 120B",
  clusterMode: "nvidia",
  enableThinking: false
});

How It Works

• Prompts are normalized into a standard message format
• Options are validated based on predefined model capabilities
• The request is streamed from the Gradient API endpoint
• Tokens are collected and cleaned
• The final message after <|message|> is returned

The function ensures output is clean of internal formatting markers.

API

gradientAI(
  prompt: string,
  options?: {
    model?: "GPT OSS 120B" | "Qwen3 235B";
    clusterMode?: "hybrid" | "nvidia"; // NVIDIA GPT only
    enableThinking?: boolean;
  }
): Promise<string>

The function always resolves with the final text response.
If streaming fails or the server output is malformed, it throws an error.

Example Output

Input: "hello AI"
Output: "Hello! How can I assist you today?"

Notes

This client intentionally abstracts from the raw stream structure.

It is suitable for CLI tools, bots, pipelines, and backend services.

Browser use depends on CORS environment.

License

MIT