NEW Google Gemma Browser AI Agent is INSANE!

The video introduces Google's newly released Gemma 4, an open-source family of AI models, and highlights a Chrome extension built by a developer named Nico Martin that brings local AI capabilities directly into the browser. The host frames the problem as the inefficiency of modern browsing — too many tabs, forgotten searches, and repetitive reading — and positions this tool as a practical solution.

Gemma 4 was released on April 2nd, 2026, and is described as Google's most capable open model to date. The 31B parameter version ranks third on the Arena AI text leaderboard among all open models globally, while the 26B version ranks sixth. Notably, the host emphasizes that Gemma 4 outperforms models 20 times its size, making frontier-level AI accessible on consumer hardware. The model is also part of a rapidly growing ecosystem Google calls the 'Gemmaverse,' with over 100,000 community-built variants already in existence.

For everyday users, the most relevant versions are the E2B and E4B — edge-optimized models designed to run on phones, laptops, and even Raspberry Pi devices. The E2B model supports a 128,000-token context window, handles images natively, and was trained on over 140 languages. These models are designed for near-zero latency and fully offline operation after initial setup.

The Chrome extension built on Gemma E2B uses the Transformers.js framework and weighs under 6 MB. Once installed, it downloads the model weights once and then operates entirely on-device. Its three core capabilities are: searching across open browser tabs to locate previously read information, querying browser history using natural language rather than exact URLs or site names, and summarizing or extracting specific information from the current web page being viewed.

The host also discusses the broader implications of Gemma 4's Apache 2.0 license, which allows commercial use, modification, and redistribution without restrictions. This openness is credited with enabling the browser extension and the wider Gemmaverse ecosystem. Real-world applications mentioned include Yale University using Gemma for cancer therapy research and a Bulgarian institution building a localized language model. Google has also confirmed that E2B and E4B models will be compatible with Android devices through an AI Core developer preview, suggesting future native on-device AI for Android users globally.

The video closes by framing Gemma 4 and the browser extension as part of a larger trend of AI moving to the edge — running locally on devices rather than in the cloud — offering faster responses, greater privacy, and reduced dependency on third-party infrastructure.