deep-chat-web-llm

A thin wrapper around the official @mlc-ai/web-llm package, repackaged for use with Deep Chat. It runs language model chats entirely in the browser with WebGPU acceleration — no server required.

This package does not contain its own implementation. src/index.ts simply re-exports @mlc-ai/web-llm:

export * from "@mlc-ai/web-llm";

All of the runtime (the MLCEngine / OpenAI-style API, the WebGPU runtime, tokenizers, etc.) comes bundled from @mlc-ai/web-llm. For the full API, model list, and documentation, see the web-llm repo.

Why this wrapper exists

It is functionally equivalent to @mlc-ai/web-llm and exists only to:

1. Provide a stable package name (deep-chat-web-llm) that Deep Chat depends on.
2. Offer a CDN <script> build that auto-attaches the library to window.webLLM for vanilla-JS use.

If you use a bundler, you can equally depend on @mlc-ai/web-llm directly.

Usage

With a bundler (assign it to the window yourself):

import * as webLLM from "deep-chat-web-llm";
window.webLLM = webLLM;

From a CDN — the bundle is an ES module, so type="module" is required (it auto-populates window.webLLM):

<script type="module" src="https://cdn.jsdelivr.net/gh/OvidijusParsiunas/web-llm@<version>/lib/index.js"></script>

Build / update

npm install @mlc-ai/web-llm@latest   # bump to the latest web-llm
npm run build                        # -> lib/index.js + lib/index.d.ts

See CLAUDE.md for the full maintainer guide: the publish flow, the CDN/window mechanics, and how the Deep Chat integration consumes this package.

License

Apache-2.0