An open-source browser agent that can verify the results
FastBrowse is an open-source browser automation tool that provides AI agents with a browser that enables them to actually interact with web pages.
Its underlying principle is “the results must be verifiable”: After completing a task, the returned answer will include references to the original text on the page, or the data will be presented in a structure defined by you. During operation, clickable elements on the page are first converted into an index, and then the model selects from this index to prevent accidental clicks; account names and passwords are only referenced by their names and are not included in the model’s prompts. In the official website demo, a complete e-commerce ordering process took 41 seconds and involved 19 steps, while the cost of developing such a model was approximately 5 cents.
It can be run locally, or using a hosted cloud browser. It supports being called within Codex and Claude Code, and is licensed under the MIT License. This tool is ideal for developers who need to perform web data collection tasks or automate business processes.