Skip to main content
ZO has three required dependencies: Claude Code CLI, Python 3.11+ with uv, and tmux. The bundled setup.sh installs them interactively if missing. 
git clone https://github.com/SamPlvs/zero-operators.git
cd zero-operators
./setup.sh
setup.sh runs eleven validation checks and offers to fix each failure interactively:
1

OS detection

Confirms macOS or Linux. Other platforms are unsupported in v1.
2

Bash version

macOS ships bash 3.2; ZO scripts are 3.2-compatible (see PR-010 for why).
3

Python 3.11+

Required for type-hint syntax used throughout src/zo/.
4

uv

Astral’s package manager. Installed via curl -LsSf https://astral.sh/uv/install.sh | sh if missing.
5

Claude CLI

Installed via curl -fsSL claude.ai/install.sh | bash if missing. The CLI handles auth on first launch.
6

tmux

Required for the conversational init/draft/build flow. Optional if you only use --no-tmux headless mode.
7

Git state

Confirms you’re inside a git repo (ZO uses git for worktree detection and commit conventions).
8

Project deps via uv sync

Resolves and installs pyproject.toml dependencies — pydantic, fastembed, click, rich, pyyaml.
9

Agent definitions

Confirms 20 agent .md files in .claude/agents/.
10

Slash commands

Confirms 24 commands in .claude/commands/.
11

zo on PATH

Symlinks .venv/bin/zo into ~/.local/bin/zo so the CLI is callable from any shell (handles conda/pyenv/system-Python users — see PR-012).
After it succeeds: 
zo --version
# cli, version 1.0.2

​
Manual install

If you prefer to install dependencies yourself: 
# Python 3.11+ — comes with macOS 14+ or via brew
brew install python@3.11

# uv — fast Python package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# tmux
brew install tmux

# Claude CLI
curl -fsSL claude.ai/install.sh | bash

# Clone and install ZO
git clone https://github.com/SamPlvs/zero-operators.git
cd zero-operators
uv sync
ln -sf "$(pwd)/.venv/bin/zo" ~/.local/bin/zo

zo --version

​
Optional: GPU support

ZO’s Docker scaffold is platform-aware:
  • Linux + NVIDIA GPU + NVIDIA Container Toolkit → emits a GPU compose template with deploy.resources.reservations.devices: capabilities: [gpu].
  • macOS → emits a CPU compose template (Docker Desktop on Mac has no GPU passthrough; for hardware acceleration, run training natively with uv run python -m src.engineering.trainer to use Apple Silicon MPS).
  • Linux without GPU → CPU compose.
Detection runs at zo init time via zo.environment.detect_environment(). Override with --gpu-host or --base-image if needed.

​
Verifying the install

After installation, run the test suite to confirm everything wires up: 
PYTHONPATH=src pytest tests/ -q
# 675 passed, 7 skipped in 7s
And run preflight on the bundled MNIST plan: 
zo preflight plans/mnist-digit-classifier.md
# 6/7 passed, 1 warnings, 0 failures.
# Ready for zo build.
If anything fails, check troubleshooting — most issues are covered by the priors documented there.

​
Next

Quickstart

Five minutes from zero to a real ML run.

Concepts

Mental model: plan, oracle, agents, gates, memory.