lociant

Lociant is an Android-native local AI runtime designed to turn low-power, always-on edge devices—such as Android phones, Rockchip boards, and small Linux systems—into controlled edge nodes. It enables local model execution, hardware-backed device capabilities, and authenticated LAN connections while keeping execution and policy on the device rather than in the cloud. The project supports importing GGUF models and running inference locally via llama.cpp. Device capabilities include screen control, accessibility features, camera access, sensors, file operations, and process management where permissions allow. External clients can connect through MCP Streamable HTTP or a control API served on port 11434. Supported device roles: - Android phone: full edge node with Flutter console, Rust backend, and Kotlin device layer - Rockchip board: headless edge node with Rust service, RKLLM NPU runtime, and terminal TUI - Linux desktop: graphical controller and development host The architecture uses a shared Rust backend for model inventory, tool registry, peer networking, and policy enforcement. GGUF inference runs as a local llama.cpp process, NCNN handles camera and vision operations on Android, and RKLLM is used on supported Rockchip boards. Tool exposure follows a cumulative permission model (read, sensor, action), and each tool has a remoteAllowed policy checked before execution. The runtime does not provide a chat UI or general-purpose agent loop; instead, it acts as a controlled capability layer that cloud or desktop clients can orchestrate. Build and run details are available in the repository's apps directory, with separate components for the Flutter UI, Rust backend, and Android device layer.