Operit AI: An Android Agent Platform That Puts the Terminal, Browser, and Device Automation in One APK
The most powerful AI agent and AI chat software on Android/Operit是一款Android上能力最为强大、发展最久的AI Agent
At a glance
- What is it?
- Operit AI is an open-source Android agent and chat app that pairs cloud or local models with deep device control, an embedded Ubuntu user space, and a visual workflow engine. It is ambitious, dense, and demanding, and it is not for casual users.
- Who is it for?
- Adopt Operit AI if you are an Android power user or developer who wants a single app that combines chat, local and cloud models, a real Linux user space, and device automation, and you accept a steep learning curve and a permission-heavy model. Do not adopt it if you need a simple chatbot or if your device is not ARM64 or runs Android below 8.0.
- Can I use it commercially?
- Yes, with conditions. LGPL-3.0 is a weak copyleft licence: you can use it inside commercial and closed-source software, but if you distribute changes to its own files, you must publish those changes under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly Kotlin, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Operit Actually Is
Operit AI is an Android application that combines an AI chat client with an agent runtime. The README describes it as a platform that connects cloud or local models to the Android system, a terminal, a browser, files, and project workspaces. It is written in Kotlin and uses Jetpack Compose, based on the repository topics. The target user is someone who wants to delegate real tasks to a model, not just exchange messages. The app supports tool calls, workflows, and extension mechanisms called ToolPkg, MCP, and Skill. It also bundles a user space for Ubuntu 24.04 ARM64, which runs through PRoot by default. That alone makes it unusual among Android AI apps. The scope is closer to a desktop development environment than a typical mobile assistant.
The Architecture: Models, Tools, and a Linux User Space
The app does not rely on one model provider. It supports OpenAI Chat and Responses, Anthropic, Gemini, and compatible services, plus local inference through MNN and GGUF models via llama.cpp. You can also connect to Ollama or LM Studio. Models are assigned per task: chat, memory, summarization, and UI control can each use a different model. Below the model layer sits a tool system. Built-in tools cover files, network, search, media, system control, and software management. Tool permissions are set in three tiers: auto-allow, ask each time, and deny. The default is ask. For UI automation, the app can use Android accessibility services, Shizuku for ADB-level permissions, or Root. The browser agent reads page structure and performs clicks, input, scrolling, and screenshots. The workspace layer provides a file tree, code editor, and preview, and it can bind to SSH, SFTP, or SAF file systems. The terminal runs an actual Ubuntu 24.04 user space under PRoot, which gives you Python, Node.js, vim, tmux, and SSH inside the app.
Getting It Running: APK, Permissions, and Local Models
Installation is not from the Play Store. You download the latest APK from the GitHub Releases page or the official website. The repository states a system requirement of Android 8.0 (API 26) or higher and ARM64 devices only. After launch, a setup guide walks you through model configuration and permission grants. For cloud models, you enter an API key or endpoint. For local models, you download model files separately; the README warns to reserve storage space according to each model's instructions. The app also offers optional features that are off by default, such as a LAN Web Chat and an HTTP API. Enabling either requires setting a Bearer Token and accepting the exposure risk. The README is explicit that the app does not host chat inference; requests go from your device to the configured provider. There is no server component to self-host for the core chat function.
The Real Limitation: Complexity and Permission Surface
The biggest limitation is not a missing feature; it is the sheer number of features and the trust each one demands. To use UI automation, you must grant accessibility access or set up Shizuku or Root. That is a serious security decision on a device that also holds your chat history, files, and credentials. The app can read your screen, simulate touches, and access files. The README acknowledges this by defaulting to an ask-permission mode and by warning against installing from unknown channels. Another limitation is platform. ARM64 and Android 8.0 exclude older devices and x86 emulators. The embedded Ubuntu user space consumes storage and memory, and the README says the footprint depends on installed tools and models. For a user who only wants a chat interface, this app is overkill and potentially risky. The learning curve is steep because the interface combines chat, a file manager, a terminal, a browser, and a workflow editor.
Extension Mechanisms: ToolPkg, MCP, Skill, and Workflows
Operit does not hard-code every capability. It provides a unified marketplace where you can install scripts, ToolPkg, Skill, and MCP servers. ToolPkg is a package format that can deliver tools, UI components, model providers, hooks, and runtime capabilities. MCP support covers local and remote servers, and it can launch via uvx or npx. Skills are separate, and each role can bind its own model, memory space, tool packages, Skills, and MCP servers. Workflows are visual graphs with nodes for triggers, execution, conditions, logic, and data extraction. Triggers include manual, scheduled, Tasker, Intent, voice, and app launch. This is a modular design, but it also means you must learn several packaging and configuration formats. The README does not provide a full API reference for ToolPkg or Skill, so expect to consult the user guide on the official site. The presence of a marketplace implies third-party code, which carries its own security considerations.
Memory, Roles, and Multi-Character Chat
The app includes a long-term memory system that stores graph-like memories, conversation history, and role cards. You can import documents, split them into chunks, edit node relationships, and search with mixed retrieval. Memory can be extracted from conversations and attachments. Roles are more than personas: each role can have its own model, memory space, tools, and skills. Multi-role group chat is supported, with @ mentions and independent histories. Role cards are compatible with Tavern JSON and PNG formats, which is a concrete interoperability point for users coming from other role-play chat apps. This feature set suggests the project grew out of an AI chat tool with strong role-play roots, then expanded into an agent platform. For a developer, the memory graph is interesting, but the documentation does not explain how the graph is stored or queried under the hood. You would need to inspect the source or the user guide to understand the schema.
Alternatives: Compare with Termux and Standalone Agent Apps
Operit occupies a space that overlaps with several other tools. The most direct comparison is Termux, which gives you a Linux environment on Android without any AI. Termux is lighter, more mature, and does not require accessibility permissions, but it has no built-in model integration or agent loop. You would pair Termux with a CLI agent like aichat or opencode. Another alternative is using a dedicated MCP client on Android, but few exist with this level of device integration. There are also cloud agent services that run on servers and only show results on your phone, but they cannot directly control your local apps or access your local files without a bridge. Operit's difference is that the agent runs on-device and has direct access to the Android UI and file system, which is both its power and its risk. If you only need a terminal, Termux is simpler. If you need remote code execution, a cloud VM with SSH might be more predictable. Operit is for those who want the agent and the environment in one app.
Maintenance and Licensing: What to Expect
The project is under the LGPL-3.0 license, which means you can link to it in a larger work, but modifications to the library itself must be released under the same license. That is relevant if you plan to fork or embed parts of the code. The repository shows a steady release cadence: v1.11.0 in May 2026, v1.12.0 in July 2026, and v1.12.1 in August 2026. The last push was September 2026, and the project is not archived. The README also points to Operit 2, a separate cross-platform rewrite using Rust for the shared runtime and Flutter for clients, with builds for Android, iOS, Windows, macOS, Linux, and Web. That is a warning sign for maintenance cost: the Android repository you are looking at may eventually be superseded. If you adopt Operit now, you should track Operit 2's progress, because the project's energy appears to be moving there. The Android version is still maintained, but its long-term future depends on how the rewrite progresses.
Editorial conclusion
Adopt Operit AI if you are an Android power user or developer who wants a single app that combines chat, local and cloud models, a real Linux user space, and device automation, and you accept a steep learning curve and a permission-heavy model. Do not adopt it if you need a simple chatbot or if your device is not ARM64 or runs Android below 8.0. Before installing, verify the exact APK source, confirm the storage footprint of the Ubuntu user space and any local models, and review the permission prompts for accessibility, Shizuku, or Root, since the app can control your UI and files. The project is actively maintained with frequent releases, but its breadth means you should test each feature you rely on, especially browser automation and workflows, on your specific device.
Community notes