Breaking

Google Sets On-Device Chrome AI Workshop at Berlin Event

June 25, 2026 at 08:00 EDT

Foundation Models
Software Dev & Coding

Google said on June 25 that it will run a hands-on workshop at its Google I/O Connect developer event in Berlin, focused on building privacy-first, on-device AI web apps that run in the browser without API keys or the cloud. The session is led by Thomas Steiner, a Google DevRel engineer specializing in Web AI and WebAssembly who has also worked on Project Fugu. (Thomas Steiner author page)

June 25 · Google I/O Connect — Berlin

Privacy-First AI, Running Entirely in Your Browser — No API Keys

A hands-on workshop on building on-device web apps powered by Gemini Nano, the small language model managed by Chrome. Inference happens locally — no cloud round-trips, no keys, no data leaving the device.

128+

Chrome version onward — via flags / origin trials

API keys & external dependencies required

100%

On-device — data & inference stay local

Cloud LLM vs. On-Device AI

Where your data goes, and what it costs to run.

Cloud LLM (API)

API key needed

Data sent externally

Per-call cost

Network latency

On-Device (Gemini Nano)

No keys

Data stays local

No call cost

No server round-trip

How it runs in the browser

Prompt API

ai.languageModel.create() / .prompt()

→

WebAssembly / WebGPU

GPU-accelerated local compute

→

Gemini Nano

Summarize · translate · rewrite · chat

22 GB+

Free disk space for model download & inference

4 GB+

GPU VRAM recommended to run the model

STRENGTHS

Privacy by default — nothing leaves the device
No keys, no per-call cost, no server latency
Seamless integration into Chrome

LIMITATIONS

Smaller model — capability ceiling vs cloud LLMs
Hardware-dependent; initial download time
No user control over which model variant runs

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.