Summon GPU instances for AI agent workflows.

Soong is a CLI tool for managing Lambda Labs GPU instances. Launch, monitor, and connect to GPU instances with a single command.

• One-command instance launch with model and GPU selection
• SSH tunneling with automatic port forwarding
• Cost controls including lease limits and idle detection
• Model registry with VRAM recommendations
• Rich terminal UI with progress indicators

# Install
cd cli && pip install -e .

# Configure (one-time)
soong configure

# Launch an instance
soong start --model deepseek-r1-70b

# Connect via SSH
soong ssh

# Start tunnel for port forwarding
soong tunnel start

# Check status
soong status

# Stop when done
soong stop

Full documentation: axiomantic.github.io/soong

• Getting Started
• User Guides
• CLI Reference
• Architecture

• Python 3.10+
• Lambda Labs account with API key

Soong includes multiple layers of cost protection:

1. Lease system - Default 4 hours, max 8 hours
2. Idle detection - Auto-shutdown after 30 minutes of inactivity
3. Cost confirmation - Shows estimated cost before launch
4. Hard timeout - 8-hour maximum instance lifetime

Contributions welcome! See CONTRIBUTING.md for guidelines.

MIT License - see LICENSE for details.