Skip to main content

​
Wafer CLI

The Wafer CLI is a powerful command-line tool for GPU kernel development, profiling, and optimization. It provides access to AI-powered assistance, performance analysis tools, and remote GPU infrastructure.

​
Installation

pip install wafer-cli
Verify the installation: 
wafer --version

​
Command Groups

The Wafer CLI organizes commands into logical groups:

Kernel Development

Commands for developing and testing GPU kernels:
  • wafer agent — AI assistant for optimization
  • wafer evaluate — Test kernel correctness
  • wafer baseline — Discover PyTorch kernel dispatch
  • wafer roofline — Performance analysis

Profiling

GPU profiling and trace analysis:
  • wafer nvidia — NVIDIA profiling tools
  • wafer amd — AMD profiling tools
  • wafer compare — Cross-platform comparison

Infrastructure

Remote GPU access and workspaces:
  • wafer workspaces — Cloud GPU workspaces
  • wafer config targets — GPU target management

Configuration

Setup and authentication:
  • wafer auth — Authentication
  • wafer config — Configuration management

​
Basic Usage

​
Get Help

# Show all commands
wafer --help

# Get help for a specific command
wafer agent --help
wafer evaluate --help

​
Common Workflows

Ask the AI agent about GPU optimization: 
wafer agent "How do I reduce bank conflicts in shared memory?"
Evaluate a kernel implementation: 
wafer evaluate gpumode --impl ./my_kernel.py --reference ./reference.py
Analyze performance against roofline model: 
wafer roofline --gpu H100 --bytes 1e9 --flops 1e12 --time-ms 0.5
Profile a kernel with NCU: 
wafer nvidia ncu analyze ./profile.ncu-rep

​
Global Options

OptionDescription
--versionShow CLI version
--helpShow help for any command

​
Environment Variables

VariableDescription
WAFER_API_KEYAPI key for Wafer services
ANTHROPIC_API_KEYDirect API key for agent (bypasses proxy)

​
Next Steps

Authentication

Set up authentication to access all features.

AI Agent

Learn to use the AI assistant for kernel optimization.

Kernel Evaluation

Test and benchmark your kernel implementations.

Workspaces

Access remote GPUs for development.