pjy010218/termux-agent-browser

termux-agent-browser solves a specific gap for people running AI agents on Android phones through Termux, a terminal app that turns Android into a Linux like environment. AI agents such as Hermes Agent need to control a web browser to navigate pages, fill in forms, take screenshots, and pull data off websites. The usual tool for this job, a project called agent-browser from Vercel Labs, only ships pre built binaries for Linux, macOS, and Windows, so it fails outright on Android's ARM64 architecture inside Termux. This project fills that gap in two ways. The simple path, called Tier 1, installs a headless copy of Chromium on the phone and exposes it through the Chrome DevTools Protocol, a standard way for tools to control a browser over a local network connection. An AI agent like Hermes can then connect directly to that endpoint and drive the browser without needing agent-browser at all. This works within about five minutes and needs no compiling. The more capable path, Tier 2, patches agent-browser's source code so it recognizes Android as a supported platform, then walks you through compiling it from source using Rust, since no prebuilt binary exists. Once built, you get agent-browser's fuller feature set on top of the same Chromium connection: accessibility tree snapshots of a page, referring to page elements by short codes to click or fill them, comparing one page state to another, and keeping track of session state across actions. The repository includes ready to run setup scripts for both tiers, a runit managed background service that keeps Chromium running and restarts it automatically, and configuration snippets for wiring the whole thing into Hermes Agent. It targets Android 13 or newer with at least 4 gigabytes of RAM, and documents version requirements for Chromium and Rust plus guidance for lower memory devices. The project is released under the Apache 2.0 license.