ClickClickClick — ИИ-агент для управления интерфейсом без API

★ 701

instavm/clickclickclick is an open-source Python framework for autonomous control of Android devices and macOS computers using any LLM, local or cloud-based. Reach for it when you need to automate something with no API: tapping through a flow in a mobile app, running repetitive routines in a desktop program, or prototyping an agent that opens apps and clicks buttons like a human. The core is a planner–finder split: one model breaks the task into steps, a second vision-capable model locates UI elements and performs taps and input; the two roles are configured independently, so you can run a cloud planner like GPT-4o or Gemini alongside a local Ollama finder to keep screenshots off external servers. Runs via CLI, a Gradio web UI, REST API, or Python API; requires Python, ADB for Android targets, and API keys for chosen models. Best results come from using Gemini Flash as both planner and finder; Ollama models are unreliable as the finder. Direction: the agent controls your device, not the other way around — run it only on your own phones and machines and supervise actions, since the model can click the wrong thing. The codebase is explicitly marked highly experimental. Good fit for U