Nanobrowser

Nanobrowser is an open-source Chrome extension that runs a multi-agent AI web agent in your browser using your own LLM API key, positioned as a flexible alternative to OpenAI Operator.

Nanobrowser is an open-source AI web agent that lives inside the Chrome browser as an extension. Instead of a hosted service with its own subscription, it lets users connect their own LLM API key and hand browsing tasks to an automated agent that acts on pages directly in the browser they already use. The project positions itself as an alternative to OpenAI Operator, with the main difference being freedom to choose which model powers the agent. The tool is built around multi-agent workflows, meaning a task is split among cooperating agents rather than handled by a single prompt loop. A user installs the extension from the Chrome Web Store, supplies credentials for a supported model provider, and then describes what should be done on the web. The agent carries out the steps in the browser, which keeps the activity inside the user's own session and environment rather than in a remote virtual machine. The homepage lists a FAQ covering supported models, getting started and how to contribute, though the answers themselves are not shown in the captured content. Nanobrowser suits developers, researchers and technically comfortable professionals who want web automation without being tied to a single model vendor. Because the code is open source, it also appeals to people who want to inspect how the agent works, change it, or contribute improvements back to the project. Bringing a personal API key means costs follow model usage from the chosen provider rather than a flat platform fee, which can be attractive for occasional use and less predictable for heavy use. Among alternatives, hosted agents such as OpenAI Operator offer a more packaged experience, while Nanobrowser trades some of that convenience for transparency, model flexibility and local control. It is limited to Chrome as a delivery surface, so it fits users who already work in that browser. Anyone considering it should expect to handle API key setup and to judge agent reliability on their own tasks.