Operit

Why we included this project

Operit runs an AI agent entirely on an Android phone. Your chat history, memory graph, model configuration, and the client itself all stay on the device, so you can either load GGUF models through MNN or llama.cpp for fully on-device inference or point it at your own cloud API endpoints with no relay service in the middle. The tool-calling layer is unusually deep for a phone app: it can grab system permissions, drive the UI through accessibility, Shizuku, or root, run an embedded browser, manage an SSH/SFTP workspace, and boot an Ubuntu user space for real terminal work. You can chain those steps into reusable visual workflows and trigger them manually, on a schedule, by voice, or from Tasker. If you want a self-contained Android assistant you can aim at your own models and extend with MCP, ToolPkg, or Skill packages, it is a serious starting point.