OpenBrowser-AI

OpenBrowser-AI provides a programmable interface that lets large‑language‑model agents control a real web browser through the Chrome DevTools Protocol. By exposing navigation, clicking, typing and data extraction as asynchronous commands, it enables agents to perform tasks such as searching, form filling or content scraping without custom browser automation code. The platform supports any LLM provider that follows the OpenAI API schema, and it ships with a persistent Python namespace for executing code alongside browser actions.

The system is built around a text‑first architecture that reduces token usage compared with other agentic browsing solutions, reporting two‑to‑six‑fold fewer tokens on benchmark suites while maintaining full task accuracy. A live VNC stream shows the browser view in real time, and the tool can be deployed via Docker or Kubernetes for production workloads. It is released under an MIT license, allowing developers to extend or integrate it with existing workflows and scheduled automation pipelines.