
Tech • IA • Crypto
A new ChatGPT Codex 5.5 Chrome extension enables direct browser control, allowing businesses to automate web-based tasks such as data extraction, research, and client outreach.
The Codex 5.5 Chrome extension allows ChatGPT to interact directly with web pages on macOS and Windows, transforming the browser into an execution environment rather than a passive tool. Users can issue commands like navigating sites, extracting data, or completing workflows in real time. This marks a shift from conversational AI toward operational automation داخل everyday tools.
The system enables task delegation such as searching listings, collecting structured data, and contacting stakeholders. In real estate, for example, agents can automate extraction of prices, energy performance certificates (EPCs), property size, and features across multiple listings. The AI can navigate pages, gather information, and compile summaries without manual browsing.
Users can create reusable “Skills”, which function as automated workflows combining prompts and browser actions. These Skills allow the AI to execute multi-step processes independently, such as opening pages, extracting data, returning results, and continuing tasks without user intervention. This introduces a modular approach to automation within ChatGPT.
The extension includes granular settings governing access to websites, browsing history, file downloads, and domain restrictions. Users can define whether actions require approval or run automatically. Sensitive data like login credentials can be stored securely using environment variables, reducing exposure while enabling automated logins.
Users can modify the Codex system prompt to define behavior, security rules, and task logic. This includes instructions for handling credentials, restricting data flows, and structuring workflows. Such customization allows organizations to standardize how the AI interacts with internal tools and external platforms.
A key development is the combination of ChatGPT Codex with Google’s NotebookLM, powered by Gemini 3.1. This setup allows the AI to query a remote knowledge base instead of loading large datasets into context. NotebookLM acts as a retrieval engine capable of processing text and images, returning relevant insights directly into ChatGPT responses.
The integration reduces reliance on locally stored retrieval-augmented generation (RAG) systems or tools like Obsidian. Instead, data can be centralized in NotebookLM, which handles indexing and retrieval. This creates a lighter, faster architecture where ChatGPT orchestrates queries while external systems manage knowledge storage.
By calling browser functions with commands such as @Chrome, users can trigger live interactions with websites. The AI can operate within active sessions, retrieve up-to-date information, and feed it back into ongoing conversations, effectively blending browsing and reasoning.
The extension effectively turns ChatGPT into a semi-autonomous agent capable of interacting with the web as a user would. It can handle navigation challenges, adapt to page structures, and execute instructions dynamically, including dealing with interface elements like prompts or consent banners.
The feature is currently limited to the United States, with users in other regions requiring workarounds such as VPN-based access to install and activate the extension. This suggests a phased rollout strategy while the technology matures.
The tool significantly reduces time spent on repetitive digital tasks, enabling automation in areas like customer service, lead generation, research, and data entry. By integrating browsing, data extraction, and response generation, it positions AI as an active participant in daily operations rather than a support tool.
The Codex Chrome extension signals a transition from conversational AI to actionable automation, enabling businesses to delegate real-world digital tasks directly to AI systems within the browser.