
OpenAI launches Atlas, an AI browser powered by ChatGPT
What's the story
OpenAI has launched a new AI-powered web browser called 'ChatGPT Atlas.' The innovative tool is built around conversational AI and integrates ChatGPT as its main interface for web navigation and interaction. The company describes Atlas as a "browser for the next era of the web," promising a more interactive and streamlined experience where users can chat with webpages, search, or assign online tasks to an AI agent.
Browser capabilities
Currently available on macOS
Atlas is now available for macOS users, with plans to expand to Windows and mobile platforms in the near future. The agent feature will be exclusive to ChatGPT Plus, Pro, and Business subscribers at first. OpenAI CEO Sam Altman said the aim is to make browsing more personal and productive by having ChatGPT learn from user preferences and "proactively find things you might want on the internet."
Design innovation
Built around the concept of 'chatting with your browser'
Ben Goodger, OpenAI's engineering lead, said Atlas was built on the concept of "chatting with your browser." Unlike traditional browsers with separate search boxes and tabs, Atlas puts ChatGPT at the center of the experience. Users can ask questions, browse sites or reference bookmarks in natural language. The interface offers all standard features like tabs, bookmarks and autofill but also comes with three unique capabilities: chat anywhere on the web, browser memory for personalization and agent mode for task automation.
Feature highlights
Browser memory for personalized experience
During a demo, OpenAI's team demonstrated how users can open a sidebar to chat with ChatGPT while viewing any webpage. For example, developers could ask ChatGPT to summarize code changes on GitHub or refine email wording within Gmail. With its new browser history feature, ChatGPT can now record the websites you visit and your activity on them to offer more personalized answers. This feature will be optional, giving users more control.
Agent mode
Agent Mode for task automation
The highlight of the demo was Agent Mode, which allows ChatGPT to take actions inside the browser such as filling forms, adding tasks to project management tools or ordering groceries online. Users can approve or monitor actions for transparency and safety. OpenAI said the agent only operates within browser tabs without access to users' local files, and memory settings can be managed or turned off entirely.