nodegpt.js

A tiny, production-minded Node.js wrapper for fast image + text generation. Simple API: generateImage(prompt, options) → Data URL (or saved file). generateReply(prompt, options) → text from a selection of Powerful models. generateReplyWithHistory(prompt, options) → text with in-memory conversation history.

Built for devs who want powerful LLM/image primitives with almost-zero ceremony — drop in, call two functions, ship features.

Install

npm install nodegpt.js

Quick start

const fs = require('fs');
const {
  generateImage,
  generateReply,
  generateReplyWithHistory,
  getConversationHistory,
  clearConversation,
  clearAllHistory
} = require('nodegpt.js');

(async () => {
  try {
    // generate image and save automatically
    const { dataUrl, saved } = await generateImage("A cyberpunk city at night", {
      width: 800,
      height: 600,
      enhance: true,
      nologo: true,
      saveTo: 'cyberpunk.png' // optional: saves raw image bytes
    });
    console.log('Saved:', saved); // 'cyberpunk.png'

    // generate text using a specific model
    const reply = await generateReply("Explain quantum computing simply.", {
      model: 'openai', // choose from the model list
      timeout: 8000
    });
    console.log(reply);

    // generate text with history persistence
    const res1 = await generateReplyWithHistory("Hello, who are you?", {
      conversationId: "chat1"
    });
    console.log(res1.text);

    const res2 = await generateReplyWithHistory("Can you continue?", {
      conversationId: "chat1"
    });
    console.log(res2.text);

    // inspect stored history
    console.log(getConversationHistory("chat1"));

    // clear history if needed
    clearConversation("chat1");
    clearAllHistory();
  } catch (err) {
    console.error('AI error:', err.message);
  }
})();

API

generateImage(prompt, options)

Generates an image for prompt and returns a Data URL (or saves to disk).

Options

• width (number) — default 512
• height (number) — default 512
• seed (number) — optional seed for reproducibility
• enhance (boolean) — default true
• timeout (ms) — default 10000 (per-endpoint)
• saveTo (string) — optional filename to write image bytes (returns { dataUrl, saved })

Returns

• If saveTo provided: { dataUrl, saved: '<filename>' }
• Otherwise: dataUrl string

generateReply(prompt, options)

Generates text for prompt using one of the supported models.

Options

• model (string) — which model to use (see list below). Default: 'openai'
• seed (number) — optional
• timeout (ms) — default 8000

Returns

• string — generated text

generateReplyWithHistory(prompt, options)

Like generateReply, but persists conversation history in memory (per Node.js process).

Options

• conversationId (string) — conversation key. If omitted, a random one is created.
• model (string) — default 'openai'
• seed (number) — optional
• contextLimit (number) — maximum total characters stored per conversation (default: 100000). If exceeded, the library will auto-start a new conversation and return { newConversationStarted: true }.

Returns

{
  conversationId: string,
  text: string,
  newConversationStarted?: boolean
}

Conversation helpers

• getConversationHistory(conversationId) → returns array of { prompt, response, ts }
• clearConversation(conversationId) → deletes history for one conversation
• clearAllHistory() → clears all conversations from memory

Models

These are the models supported — use their name in generateReply(..., { model: '<name>' }).

| name | short description | input → output | | ------------- | ------------------------------------: | -------------- | | gemini | Gemini 2.5 Flash Lite (api.navy) | text → text | | mistral | Mistral Small 3.1 24B | text → text | | nova-fast | Amazon Nova Micro | text → text | | openai | OpenAI GPT-5 Nano | text → text | | openai-fast | OpenAI GPT-4.1 Nano (fast) | text → text | | qwen-coder | Qwen 2.5 Coder 32B (coder-focused) | text → text | | bidara | BIDARA (NASA research assistant) | text→ text | | midijourney | MIDIjourney (creative text2text tool) | text → text |

Best practices

• Wrap calls in try/catch — network or DNS issues can still happen.
• Use timeout option for responsive UX.
• For production high-availability, use saveTo for images if you want to persist results server-side.
• For multi-user apps, consider mapping your own user IDs → conversationIds so history stays separate.
• History is ephemeral (in-memory only). If the process restarts, history resets.

Error handling example

(async () => {
  try {
    const img = await generateImage("A blue whale made of glass", { timeout: 7000 });
    // use img...
  } catch (err) {
    console.error('generateImage failed:', err.message);
    // fallback logic or user-friendly message
  }
})();

Security & privacy

• The package hides target endpoints in its internal implementation (for a clean API surface). If you need full transparency about upstream hosts, ask and we’ll enable a verbose/debug mode.
• History is stored only in process memory, never persisted unless you explicitly save outputs via saveTo.

Contributing

Am working on a social media app, email [email protected] to join me

License & Author

MIT Author: Ismail Gidado & Textmob