Mobilerun

Mobilerun is an open source framework for building AI agents that control iOS and Android devices. A developer states a goal in natural language, and the agent operates the phone to complete it, reading the screen through the accessibility tree with a screenshot based vision mode as a fallback. The framework is installed with pip, is MIT licensed, and works with any language model, including Gemini, OpenAI, Anthropic, DeepSeek, Qwen, Ollama, OpenRouter and any OpenAI compatible endpoint. The framework is aimed at developers who want to automate tasks across mobile apps, write mobile tests in plain English without Appium or XPath, or extract structured data from apps that have no public API. It supports multi step reasoning, structured output through Pydantic models and a multi agent design. The vendor publishes benchmark results and recorded action trajectories for iOSWorld and AndroidWorld. Alongside the framework it sells a hosted cloud phone platform and a hosted agent called mobilerun VA, which is in beta. The framework itself is free. Cloud plans are priced monthly. Free costs $0 and includes one concurrent phone, 60 device minutes and 500 credits. Developer costs $20 per month with 5 concurrent phones, 600 device minutes and 1,500 credits. Startup costs $99 per month with 25 concurrent phones and 3,000 device minutes. Scaleup costs $299 per month with 100 concurrent phones. Extra device minutes cost $0.03 each on paid plans. Enterprise plans are arranged through a call.