Try it in two minutes

Mac and Linux, no Rust needed:

curl -fsSL https://zerd.si/install.sh | sh

Windows 10/11 (PowerShell), no Rust needed:

irm https://zerd.si/install.ps1 | iex

It installs zerd.exe into %LOCALAPPDATA%\zerd\bin and adds that folder to your user PATH (open a new terminal after). Or grab the tarball / zip for your machine from the releases page. Building from source (any system):

git clone https://github.com/Adityaadpandey/zerd.si && cd zerd.si
cargo build --release                                       # binary: target/release/zerd
ln -sf "$PWD/target/release/zerd" ~/.local/bin/zerd       # put it on your PATH (Windows: target\release\zerd.exe)

On Windows, install Git for Windows first. zerd needs git anyway, and it ships the bash that zerd runs shell commands in, because models write bash. zerd looks for that bash next to your git (not the WSL stub in System32, which runs in a Linux VM and can't see your Windows paths). Without it zerd falls back to PowerShell, and commands a model wrote for bash will often fail. See Windows notes.

Give it a key. Either a .env with API_KEY=... (in the folder you run it from, or in zerd's data folder so it works from anywhere) or, if you already use the Codex CLI, nothing at all: -m codex/<model> reuses that login. To use OpenAI, Anthropic, Gemini, Groq, OpenRouter, Ollama or another provider directly, start zerd and type /connect (see Models and routing). The default model is MiniMaxAI/MiniMax-M3; change it with -m <model> or /model.

cd any/project && zerd                                     # interactive, full-screen
zerd -m claude-haiku-5-5 --approval auto "fix the failing test in tests/test_paging.py"   # one shot
zerd --continue                                            # pick up the last session in this folder

First things to type once you're in: /help, then try review my changes, /cost to see where tokens went, /undo if you don't like what it did.