New Release

Amplifier in
Your Browser

Introducing webruntime and webllm bundles for fully serverless AI

Experimental
January 2026
The Vision

What if Amplifier could run
entirely in the browser?

No servers. No API keys. No cloud costs.
Just open a webpage and start building with AI.

Announcing

Two new bundles

amplifier-bundle-webruntime
Run Amplifier sessions in web environments using Pyodide (Python in WebAssembly). Includes the webruntime-developer agent for building browser apps.
amplifier-bundle-webllm
WebLLM integration for local GPU inference via WebGPU. Enables fully serverless AI that runs 100% in the browser.
How It Works

Real Amplifier in the browser

Your Web Browser
↓
Pyodide (Python in WebAssembly)
↓
amplifier-core (real kernel)
↓
WebGPU Provider
↓
WebLLM + Local Model

Not a simulation — the actual Amplifier framework

Why This Matters

Serverless AI benefits

🔒
Complete Privacy
Data never leaves the user's device. All inference happens locally.
💰
$0 API Costs
No per-request charges. Runs entirely on the user's GPU.
✈️
Works Offline
After initial model download, works without internet.
📦
Single HTML File
Embedded wheel pattern for self-contained apps.
Featured Agent

webruntime-developer

An agent that builds browser-based Amplifier applications

🏗️
Builds Apps
Creates complete HTML apps with embedded Pyodide + WebLLM
🧪
Tests Autonomously
Uses Playwright to verify apps work correctly
🔧
Iterates
Fixes issues and re-tests until working
Get Started

Add the bundles

# Add the webllm bundle (includes webruntime) amplifier bundle add git+https://github.com/microsoft/amplifier-bundle-webllm # Or just webruntime if using a different provider amplifier bundle add git+https://github.com/microsoft/amplifier-bundle-webruntime # Use the webruntime-developer agent amplifier run "build a browser chat app with WebLLM"
Sources

Research Methodology

Data as of: February 20, 2026

Feature status: Experimental

Repositories:

Research performed:

Technology stack: Pyodide v0.27.0, WebLLM v0.2.79, amplifier-core wheel (embedded)

Primary contributor: Diego Colombo

Gaps: Inference speed benchmarks vary by GPU hardware; model quality is limited by WebGPU-compatible quantized models. No independent line count verification (bundles not cloned locally).

See It Live

Try it yourself

This presentation includes a live demo running real amplifier-core in your browser right now.

Click "Try the Demo" at the end to chat with Amplifier
More Amplifier Stories

Try the Demo

Real amplifier-core via Pyodide + WebGPU

Browser Bundles Demo Loading...
Initializing Amplifier...
○
Check WebGPU
○
Load Python Runtime
○
Download LLM Model
○
Start Amplifier Session