The privacy-protecting AI browser Eclipse has been released... without going through a central server.

The AI web browser that prioritizes user privacy has officially been commercialized. The new product “Eclipse” from Estonian startup Sigma Browser features a large AI model built directly locally, without transmitting user data to the cloud. Recently, major browsers have been competing to integrate AI, and this product offers an alternative focused on privacy protection, standing out in the tech market.

Eclipse enables all activities such as search history, chat content, and document analysis to be processed internally on the device by directly deploying large language models that can run in offline environments on PCs. Nick Trenkler, co-founder of Sigma, emphasized in the release: “AI is becoming increasingly centralized and costly,” and “users should be able to utilize advanced AI without sacrificing privacy or incurring ongoing cloud fees.”

Most of the major browsers currently utilize AI in a cloud-based manner. Google's Chrome integrates Gemini, while Mozilla has integrated its own AI model into Firefox. Additionally, Perplexity AI's Comet and OpenAI's Atlas are also browsers that rely on cloud access. Throughout this process, users' queries are sent to AI servers for processing, raising ongoing concerns about data leakage and privacy infringement.

The differentiation of Eclipse lies in its comprehensive blockage of this process. Sigma indicates that it is fundamentally equipped with an AI model without filtering restrictions, which also eliminates the possibility of information distortion due to political or ideological biases. In addition, the product also includes a feature to directly analyze PDF documents on PC, further enhancing its practicality in local environments.

Of course, to smoothly run such locally-based models, a certain level of hardware performance is required. Typically, to run a medium AI model with 7 billion parameters, at least 16-32GB of memory and a graphics card at the level of NVIDIA RTX 3060 is needed as a basic configuration. Sigma recommends using RTX 4090 or higher specifications.

Browsers that provide a fully local operating environment are not new, and Eclipse is not the first. Previously, Brave had included a feature in its “Leo” AI function that allowed users to install and run learning models on their own. However, some pointed out that the installation process has a high technical threshold. In contrast, Eclipse is considered to have lowered the usage threshold to the greatest extent because users can use the basic built-in model directly without any additional setup.

With this product launch, Sigma announces its shift towards a privacy-centric “user-controlled AI” direction. As AI technology advances and the limitations of centralized processing structures are highlighted, the likelihood of a future market trend shifting towards services where users can directly control their AI usage environment and ensure data sovereignty is increasingly growing.