coding agent for GPU-kernel development

Built by Herdora

Quick install

npm install -g herdora
export OPENAI_API_KEY=sk-…

Launch

cd /your/project
herdora

A framed chat prompt appears:

┌─ Herdora ─────────────────────────────────────────────-┐
│ You’re chatting with herdora                           │
│ Type !help for commands; Ctrl-C twice to exit          │
└────────────────────────────────────────────────────────┘
herdora ›

You're ready to optimize your GPUs!

Example session

herdora › read kernels/matmul.cu
(🥷 Herdora streams the file with syntax highlighting)

herdora › optimise this kernel for H100, target 128x128 tiles
(⚙️  Herdora shows a diff — you confirm Y/n)

herdora › what speed-up should I expect vs PyTorch matmul on 4096×4096?
(💡 Explanation + back-of-envelope FLOP math)

herdora › create tests/matmul_benchmark.py to time both versions
(✏️  New file drafted, pending confirmation)

All operations—read, write, diff—are driven by free-form chat; no separate sub-commands.

Development

To develop and test the CLI locally:

1. Clone the repository

   git clone https://github.com/yourusername/herdora-cli.git
   cd herdora-cli

2. Install dependencies

   npm install

3. Build the TypeScript code

   npm run build

4. Run in development mode

   npm run dev

5. Link the package locally

   sudo npm link

Now you can use the herdora command from anywhere on your system.