What it does Operit is a Kotlin-based Android AI assistant that tries to do everything: local LLM inference via MNN/llama.cpp, cloud API chat, voice interaction, tool calling, memory management, and UI automation. It also bundles a full Ubuntu 24 chroot environment with apt, Python, Node.js, and vim. The pitch is a “fully independent” mobile AI agent that doesn’t need a server—except when it does for API calls.

The interesting bit The UI automation goes deeper than most mobile AI toys: it supports three privilege escalation paths (accessibility, ADB, root) and can manipulate virtual displays. The AutoGLM-style agent can literally tap buttons on your behalf. Whether this is useful or terrifying depends on your threat model.

Key highlights

• Local GGUF model support through MNN and llama.cpp for offline inference
• 40+ built-in tools plus MCP/Skill plugin marketplace with uvx/npx support
• Ubuntu 24 environment with full package management, SSH, chroot, tmux
• Three-channel UI automation: accessibility, ADB, and root with virtual display support
• Role cards with independent chat histories, QR-code sharing, and group chat between personas
• Floating window mode, WebP desk pet, token usage stats, and enough theme knobs to keep you busy for hours

Caveats

• The README claims “completely independent operation” but immediately qualifies it with “except API calls”—so cloud-dependent features are still cloud-dependent
• Requires Android 8.0+, 4GB+ RAM recommended; local LLMs on phones remain a battery and thermal challenge
• The feature surface is enormous (browser automation, APK tool, Java bridge, xAI video…) which suggests maintenance burden; the changelog is already dense with bugfix categories

Verdict Grab it if you want a Swiss Army knife AI agent on Android and don’t mind living in a Chinese-language-first ecosystem. Skip if you just need a clean chat interface—this is emphatically not that.