GenericAgent

Why we included this project

GenericAgent gives an LLM direct system-level control over a local machine, including the browser, terminal, filesystem, keyboard and mouse, screen vision, and Android devices over ADB. The whole framework is small: about 3,000 lines of seed code, nine atomic tools, and a minimal agent loop. Each time it solves a task, it records the execution path as a reusable skill, so its abilities accumulate with use and it stops re-reading the same prompts on repeat work, which keeps token usage down. Teams that want a lightweight, hackable base for desktop and browser automation, and are fine letting the agent grow its own procedures instead of shipping a big pre-built skill library, will get the most out of it. The self-evolution loop itself is small enough to read end to end, which also makes this a good starting point for anyone studying how such agents are built.