CogniSelect SDK

Browser SDK for CogniSelect – context-menu AI features.

Demo

You can try it out on CogniSelect.

Features

• Contextual Actions: Generate summaries, quizzes, and more based on selected text.
• Generate your own text on areas/inputs: Create custom text generation actions.
• Prebuilt Models: Use preconfigured models for various tasks.
• Customizable: Add your own actions and models.
• Persistent Cache: Store models in browser storage for faster access.
• Lightweight: Minimal footprint, designed for in-browser use.
• Easy Integration: Simple API for quick setup and customization.
• Cross-Browser Support: Works on all modern browsers.
• Open Source: Available under the Apache 2.0 License.
• Community Driven: Contributions and feedback are welcome.

Installation

npm

Install via npm:

npm install @cogniselect/sdk

CDN

ES Modules

Load as an ES module:

<script type="module">
  import { CogniSelect } from 'https://cdn.jsdelivr.net/npm/@cogniselect/sdk/dist/index.esm.js';
  // or unpkg:
  // import { CogniSelect } from 'https://unpkg.com/@cogniselect/sdk/dist/index.esm.js';
</script>

UMD Build

<script src="https://cdn.jsdelivr.net/npm/@cogniselect/sdk/dist/index.umd.js"></script>
<script>
  const { CogniSelect } = window;
</script>

Usage

Import and initialize CogniSelect:

import { CogniSelect, prebuiltModelIds } from '@cogniselect/sdk';

const cogniselect = new CogniSelect({
  model: prebuiltModelIds[0], // or 'gemma-2-2b-it-q4f16_1-MLC',
  cacheStrategy: 'persistent', // Cache model in browser storage
  actions: [
    {
      category: 'summarize',
      description: 'Summarize the selected text',
      prompt: 'Summarize the selected text: "{text}"',
    },
    {
      category: 'quiz',
      description: 'Generate a quiz based on the selected text',
      prompt: 'Generate a quiz based on the selected text: "{text}", minimum 5 questions. Highlight the correct answers.',
    },
    // more actions...
  ],
  // optional: override default model list
  // modelList: prebuiltModelIds,
});

// You have to attach the CogniSelect instance to a DOM element
cogniselect.attachCogniSelectStatus('#model-indicator-container');

Available Models

Use the prebuiltModelIds array to see all supported models:

import { prebuiltModelIds } from '@cogniselect/sdk';
console.log(prebuiltModelIds);

Development

Build the project:

npm run build

Acknowledgements

This project would not be possible without the excellent work of mlc-ai/web-llm, which provides the core in-browser LLM engine. We gratefully acknowledge their contribution.

License

Apache 2.0 License. See LICENSE.md.