AAswordman/Operit

Overview

Operit AI is an open-source Android AI agent platform distributed as an Android app. It connects cloud models, custom OpenAI-compatible endpoints, Ollama/LM Studio, or on-device MNN and GGUF models to tools that act on the phone: files, search, media, the system, a built-in browser, a terminal, and project workspaces. Tasks that would normally require clicking through apps or running commands can therefore be delegated to a model with tool calling, workflows, MCP services, Skills, or ToolPkg extensions.

The app also doubles as a chat client with attachments, message branching, character cards, multi-character group chats, graph-based long-term memory, voice input/output, and avatars. This repository is the Android edition; the cross-platform successor is a separate Rust/Flutter project, Operit 2.

Quick start

1. Check the device: Android 8.0 (API 26) or newer, ARM64 (arm64-v8a) only (requirements).
2. Download the latest APK from the Releases page or the Operit website. The README warns against packages from unknown sources.
3. Install and launch the app.
4. Follow the setup flow to configure a model and grant permissions.
5. Start a task or chat. Android automation requires Accessibility, Shizuku, or Root permissions depending on the chosen approach; cloud model requests go from the device to the endpoint you configure.

Tutorials and examples live on the Operit user guide.

Constraints and data boundaries

• Platform: Android 8.0+ on ARM64 only, so x86 emulators or 32-bit devices are not covered.
• Resources: memory and storage use depends on the terminal environment, installed tool packages, and local models; size accordingly.
• Permissions: device automation, virtual displays, browser control, and similar features rely on Accessibility, Shizuku, or Root access where applicable.
• Local storage: chats, character cards, memories, and model configurations are stored locally; cloud inference is sent to the provider you configure.
• Network exposure: the optional LAN Web Chat and HTTP API are disabled by default and should be enabled only after reviewing bearer-token and exposure risks.
• Licensing: the main code is LGPL-3.0-only, while tools, templates, examples, and third-party dependencies may carry other licenses (license notes).

HotRepo's take

HotRepo has no daily ranking or star-curve snapshot to report for this repository, so there is no trend to cite here. What the material does show is a long, documented release history: the changelog traces the change from basic AI chat and Shizuku/Root integration in v1.0.0 (April 2025) to a workflow engine, Ubuntu 24.04 user space, GGUF local inference, a browser agent, and a unified marketplace by v1.12.0 (release summaries). The scope is unusually broad for a single Android app, which makes the setup cost and permission surface the main things to weigh before adopting it.