AIOPE — AI-First Operations

An Android IDE that combines a code editor, multi-shell terminal, and AI chat with agentic tool use — all in one app. LLM traffic routes through litellm on-device via embedded CPython 3.11, giving native access to 20+ providers with no cloud middleman.

Features

Code Editor

• Sora Editor with TextMate syntax highlighting (Python, Kotlin, Java, JS, JSON, XML, Shell, Markdown)
• File browser, file tree sidebar, line numbers, word wrap
• Multi-language code formatter with options dialog
• "Code Assist" button sends editor content to AI chat

Terminal

• sh — Android shell via /system/bin/sh
• proot Ubuntu — Full Debian userland via proot (apt, node, python, git)
• rish — Elevated Shizuku shell for system-level access
• SSH — Remote connections via JSch
• Tab management for multiple sessions

AI Chat

• Streaming responses from any of 20+ LLM providers
• 11-agent team with specialized roles (architect, dev, QA, security, etc.)
• Agentic tool loop (up to 6 rounds): read_file, write_file, list_files, search_code, run_shell, fetch_url
• Multi-shell agent tools — AI can run commands in sh, proot Ubuntu, or rish
• Native + XML fallback tool calling for models without function calling support
• Reasoning model support (o1/o3/o4) with separate reasoning display
• Multimodal: images, camera, screenshots (PixelCopy), audio, speech-to-text
• On-device OCR via ML Kit
• Conversation management, export, auto-titling
• Gems library for saving snippets, images, OCR results
• Problem Library for local interaction search

MCP (Model Context Protocol)

• 4 built-in servers with native Kotlin transport (no Node.js required):
  • Filesystem — read, write, list, search, move files, get metadata
  • Git — status, log, diff
  • Fetch — HTTP GET with configurable length
  • SQLite — read-only SQL queries against any .db file
• User-added servers via STDIO, SSE, WebSocket, or HTTP transports
• Server management UI

LLM Providers

20+ built-in providers with dynamic discovery of 90+ more via litellm's cost DB:

OpenAI · Anthropic · Google AI Studio · Google Vertex AI · AWS Bedrock (SSO + IAM) · Azure OpenAI · Groq · Mistral · Together AI · DeepSeek · OpenRouter · NVIDIA NIM · GitHub Models · Cloudflare AI · xAI · Moonshot · Alibaba Qwen · Snowflake Cortex · Ollama · LM Studio · llama.cpp · Custom OpenAI-compatible

• Three-tier model system: custom models + live-fetched + static defaults
• Per-provider profiles with temperature, top-p, top-k, max tokens, reasoning effort
• Bedrock SSO with full OIDC device auth flow

Extensibility

• Skills — 11 built-in modular LLM capability bundles (code gen, analysis, testing, docs, security, etc.)
• Hooks — 11 built-in event-driven automations (auto code review, error explainer, commit helper, etc.)
• Plugins — Typed extension framework with 14 extension points

Architecture

MainActivity (3-tab ViewPager2)
    ├── EditorFragment      → Sora Editor + TextMate
    ├── TerminalFragment    → sh / proot / rish / SSH
    └── AIChatFragment      → LLMOrchestrator
                                  ├── Python (litellm via Chaquopy)
                                  ├── AgentManager (11 agents)
                                  ├── MCPManager (native + remote servers)
                                  └── ChatRepository (Room DB)

Tech Stack

Layer	Technology
Language	Kotlin 1.9.20
Embedded scripting	Python 3.11 via Chaquopy 15.0.1
LLM gateway	litellm 1.82.4
Code editor	Sora Editor 0.22.0 + TextMate
Terminal	Termux terminal emulator v0.118.0
Database	AndroidX Room 2.6.1
OCR	ML Kit Text Recognition 16.0.0
Markdown	Markwon 4.6.2
Elevated perms	Shizuku 13.1.5
SSH	JSch 0.1.55
Min SDK	26 (Android 8.0)
Target SDK	34 (Android 14)

Building

Requires JDK 17, Android SDK 34, and the Chaquopy Gradle plugin.

# Docker build (recommended)
docker exec xnet-dev bash -c 'cd /workspace/aiope && ./gradlew :app:assembleDebug --parallel --build-cache'

# Or use build.sh
./build.sh              # full debug build
./build.sh --quick      # fast iteration (no PNG crunch)
./build.sh --clean      # clean + full build
./build.sh --install    # build + adb install

# Install
adb install -r app/build/outputs/apk/debug/app-debug.apk

A post-build pipeline patches litellm .imy archives (patch_litellm_imy.py) and injects pre-compiled Rust .so extensions for pydantic_core, jiter, rpds, and fastuuid (inject_native_imy.py).

Configuration

Add API keys per provider in Settings → Provider Profiles. Bedrock uses SSO device auth (no keys needed). Local providers (Ollama, LM Studio, llama.cpp) need only a base URL.

License

AGPL-3.0 — See LICENSE file