AI-powered natural language control center — one mobile interface to rule everything.
English | 中文
AImpire is a mobile-first platform that lets you control complex systems through natural language conversation. Instead of learning CLI tools, dashboards, or APIs for every system you manage, you talk to an AI agent that understands your intent and executes the right actions.
The interface is a single PWA chat app (installable on iOS/Android). The backend is a Claude-powered agent that routes your message to the appropriate feature module, executes tools, and streams results back in real time.
┌─────────────────────────────────────────────────┐
│ Mobile PWA (iOS / Android) │
│ Natural language · Streaming │
│ ┌───────────────────────────────────────────┐ │
│ │ Mode: [GPU Server] [PC] [Web] [Data] … │ │
│ └───────────────────────────────────────────┘ │
└─────────────────────────┬───────────────────────┘
│ HTTPS + Bearer Token
▼
┌─────────────────────────────────────────────────┐
│ Cloud Server (FastAPI) │
│ │
│ Claude API (streaming) · Multi-session │
│ ┌────────────────────────────────────────┐ │
│ │ Feature Router │ │
│ │ gpu_server │ pc_control │ web │ … │ │
│ └────────────────────────────────────────┘ │
│ Core Agent: tool use · history · persistence │
└──────┬───────────────────────┬───────────────────┘
│ SSH │ Future protocols
▼ ▼
┌─────────────┐ ┌─────────────────┐
│ GPU Server │ │ Your PC / APIs │
└─────────────┘ └─────────────────┘
| Mode | Status | Description |
|---|---|---|
| 🖥️ GPU Server | ✅ Available | Manage remote GPU servers — run experiments, monitor jobs, sync code, analyze logs |
| 💻 PC Control | 🔜 Coming soon | Control your local PC — automate workflows, manage files, run scripts |
| 🌐 Web Automation | 🔜 Coming soon | Browser automation — data scraping, form filling, web interactions |
| 📊 Data Analysis | 🔜 Coming soon | Analyze datasets — generate charts, run statistics, export reports |
Each feature is an independent module. You can enable only what you need, or build your own.
git clone https://github.com/szy233/AImpire.git
cd AImpirepip install -r requirements.txtcp configs/config.example.yaml configs/config.yamlFill in configs/config.yaml — at minimum you need:
claude:
api_key: "sk-ant-..." # Your Anthropic API key
server:
auth_token: "your-token" # Secret for mobile auth (anything you choose)See Config Reference for all options.
python setup_vapid.pyCopy the output into config.yaml under server.vapid_public_key and server.vapid_private_key.
python generate_icons.pyuvicorn web.api_server:app --host 0.0.0.0 --port 8000Navigate to http://YOUR_SERVER_IP:8000 in Safari / Chrome, then:
- iOS: Share → Add to Home Screen (full PWA experience)
- Android: Menu → Install App
On first launch, enter your auth_token in the settings dialog.
AImpire/
├── features/ # Feature modules (one per capability)
│ ├── gpu_server/ # ✅ GPU server management
│ │ └── README.md # Feature docs, tools, setup
│ ├── pc_control/ # 🔜 PC control (planned)
│ ├── web_automation/ # 🔜 Web automation (planned)
│ └── data_analysis/ # 🔜 Data analysis (planned)
│
├── core/ # Shared infrastructure
│ ├── agent.py # Core agent: Claude API, tool dispatch, history
│ ├── state_manager.py # Experiment/task state (SQLite)
│ └── project_manager.py # Project config (.agent.yaml)
│
├── tools/ # Tool implementations
│ ├── ssh_executor.py # Remote SSH execution
│ ├── git_manager.py # Git operations + code sync
│ ├── file_sync.py # File transfer (SCP)
│ └── log_analyzer.py # Training log parsing
│
├── web/ # Frontend + API server
│ ├── chat.html # Mobile PWA — single-page app
│ ├── api_server.py # FastAPI: auth, streaming, push, sessions
│ ├── manifest.json # PWA manifest
│ └── sw.js # Service Worker (cache + push events)
│
├── configs/
│ ├── config.example.yaml # Template (copy → config.yaml)
│ └── config_manager.py # Pydantic config loader
│
├── setup_vapid.py # VAPID key generator
├── generate_icons.py # PWA icon generator
└── requirements.txt
See CONTRIBUTING.md for the full guide. The short version:
- Create
features/<your_feature>/ - Implement tools in
core/agent.py(or a new agent subclass) - Add the mode entry to the
MODESarray inweb/chat.html - Set
available: truewhen ready
# ===== GPU Server =====
gpu_server:
host: "localhost" # "localhost" for tunnel mode, or direct IP
port: 22
username: "your_user"
key_path: "~/.ssh/id_ed25519"
password: null # Leave null to use SSH key
tunnel_port: 2222 # Reverse tunnel port; null = direct SSH
workspace: "/home/user/projects"
conda_env: null # Conda env to activate
# ===== Local / Cloud Server =====
local:
workspace: "./workspace"
results_dir: "./results"
db_path: "./data/state.db"
# ===== Claude API =====
claude:
api_key: "sk-ant-xxxxx"
model: "claude-sonnet-4-20250514"
max_tokens: 4096
# ===== Web Server =====
server:
host: "0.0.0.0"
port: 8000
auth_token: "your-secret-token"
vapid_public_key: "" # From setup_vapid.py
vapid_private_key: "" # From setup_vapid.py
# ===== Notifications (optional) =====
notify:
telegram_bot_token: null
telegram_chat_id: nullOn the GPU server:
autossh -M 0 -N -R 2222:localhost:22 \
-o ServerAliveInterval=60 \
YOUR_USER@YOUR_CLOUD_SERVER_IPOr as a systemd service — see features/gpu_server/README.md.
Then in config.yaml:
gpu_server:
host: "localhost"
tunnel_port: 2222[Unit]
Description=AImpire
After=network.target
[Service]
User=YOUR_USER
WorkingDirectory=/path/to/AImpire
ExecStart=/path/to/venv/bin/uvicorn web.api_server:app --host 0.0.0.0 --port 8000
Restart=on-failure
RestartSec=5
[Install]
WantedBy=multi-user.targetMIT
AImpire 是一个移动端优先的 AI 控制平台。通过自然语言与 Claude 对话，即可管理 GPU 服务器、操控本地 PC、执行自动化任务——无需记忆命令行语法，一个聊天界面搞定一切。
当前已实现的功能模块：GPU 服务器管理（运行训练、监控任务、同步代码、分析日志）。
更多模块（PC 操控、Web 自动化、数据分析）正在开发中。各模块相互独立，可按需启用，也可自行扩展。
详见各功能目录下的 README：features/gpu_server/README.md