UCAI SDK · JavaScript

A ~3 kB, zero-dependency wrapper that lets any web or Node app talk to an OpenAI-compatible relay (for example https://ucai.ngrok.app/api/v1).

It supports everything in the modern spec:

• streaming chat
• tool (function) calls
• JSON-schema responses
• plugins (web search, PDF parsing, …)
• multimodal images & PDFs
• image generation (dall-e-2/3, gpt-image-1)

…and ships helpers so you never hand-craft request JSON again.

1 · Quick start

npm i ucai-sdk          # or drop dist/ucai.iife.js in a <script>

import { UCAI } from 'ucai-sdk/dist/ucai.esm.js';

const ai = new UCAI({
  baseUrl: 'https://ucai.ngrok.app/api/v1',
  apiKey : 'sk-YOUR-KEY',
});

Zero-build via CDN

<script type="module">
  import { UCAI } from 'https://cdn.jsdelivr.net/npm/[email protected]/dist/ucai.esm.js';
  const ai = new UCAI({ baseUrl:'…', apiKey:'…' });
</script>

ES5 global

<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/ucai.iife.js"></script>
<script>
  const ai = new UCAI.UCAI({ baseUrl:'…', apiKey:'…' });
</script>

2 · Core calls

2.1 Plain chat

const res = await ai.chat({
  messages: [{ role:'user', content:'Summarise tokenisation in one line' }],
});
console.log(res.choices[0].message.content);

2.2 Streaming chat

const { stream } = await ai.chat({
  stream: true,
  messages: [{ role:'user', content:'Write a haiku about code' }],
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices?.[0]?.delta?.content || '');
}

2.3 Abort mid-stream

const { stream, controller } = await ai.chat({ …, stream:true });
setTimeout(() => controller.abort(), 2000);

3 · Tool calling (functions)

const tools = [
  {
    schema: {
      name: 'getPi',
      description: 'Returns π with given decimals',
      parameters: {
        type: 'object',
        properties: { d: { type:'integer', minimum:1, maximum:10 } },
        required: ['d'],
      },
    },
    func: ({ d }) => +Math.PI.toFixed(d),
  },
];

const out = await ai.chat({
  messages: [
    { role:'system', content:'You can call getPi' },
    { role:'user',   content:'Give me pi with 5 decimals' },
  ],
  tools,
});
console.log(out.choices[0].message.content);

4 · Structured JSON answers

const schema = {
  name: 'projectIdea',
  strict: true,
  schema: {
    type: 'object',
    properties: {
      idea:       { type:'string' },
      difficulty: { type:'integer', minimum:1, maximum:10 },
    },
    required: ['idea','difficulty'],
  },
};

const out = await ai.chat({
  messages: [{ role:'user', content:'Side-project idea in JSON' }],
  schema,
  validateSchema: true,          // optional runtime AJV check
});
console.log(out.parsed);         // already JSON.parse’d

5 · Plugins (server-side helpers)

import { PLUGIN } from 'ucai-sdk/dist/ucai.esm.js';

const news = await ai.chat({
  messages: [{ role:'user', content:'Latest news on Google I/O 2025?' }],
  plugins : [PLUGIN.web],        // web-search plugin
});
console.log(news.choices[0].message.content);

Combine freely:

plugins : [PLUGIN.web, PLUGIN.pdf('mistral-ocr')]

('pdf-text' is the free default engine; 'mistral-ocr' handles scanned docs.)

6 · Images & PDFs as input

import {
  attachImageUrl,
  attachPDFBlob,
} from 'ucai-sdk/dist/ucai.esm.js';

/* image url */
await ai.chat({
  messages: [{
    role:'user',
    content:[
      { type:'text', text:"What's in this image?" },
      attachImageUrl('https://example.com/photo.jpg'),
    ],
  }],
});

/* PDF blob (e.g. <input type=file>) */
const pdfPart = await attachPDFBlob(fileInput.files[0]);
await ai.chat({
  messages:[{ role:'user', content:'Summarise this PDF' }],
  files:[pdfPart],
  plugins:[PLUGIN.pdf()],          // use OCR engine if needed
});

7 · Image generation

7.1 URL response (dall-e-2/3)

const res = await ai.generateImage({
  prompt: 'Singapore skyline in synth-wave style',
  model : 'openai/dall-e-3',
  size  : '1024x1024',
});
document.getElementById('out').src = res.data[0].url;   // expires in 60 min

7.2 Base-64 response (gpt-image-1)

const res = await ai.generateImage({
  prompt: 'Cute corgi in space suit',
  model : 'gpt-image-1',
  response_format: 'b64_json',      // default for this model
});

const img = new Image();
img.src = `data:image/png;base64,${res.data[0].b64_json}`;
document.body.append(img);

8 · Helper cheat-sheet

| Helper | Purpose | | ----------------------------------- | --------------------------------- | | attachImageUrl(url) | Image by public URL | | attachImageBlob(blob) | Local file/Blob image | | attachPDFBlob(blob) | Local PDF attachment | | toDataURL(buf,mime) | Low-level base-64 utility | | PLUGIN.web / PLUGIN.pdf(engine) | Plugin presets | | compressSchema(schema) | Minify JSON-schema before sending | | validateAgainst(schema,obj) | Runtime AJV validation (async) |

9 · One-file bundle

npm i            # installs esbuild (dev dep)
npm run build    # → dist/ucai.esm.js & ucai.iife.js

Drop ucai.iife.js into any legacy HTML page:

<script src="ucai.iife.js"></script>
<script>
  const ai = new UCAI.UCAI({ baseUrl:'https://…', apiKey:'sk-…' });
  ai.chat({ messages:[{role:'user', content:'hi'}] }).then(console.log);
</script>

10 · Full demo

demo/demo.js exercises every feature—non-stream, streaming, tools, schema, plugins, multimodal, and image generation.

node demo/demo.js           # needs Node ≥ 18 for global fetch

Enjoy the speed boost—one import, all modern OpenAI goodies.