Start Here

Welcome! This guide is written for people who just want to get things done with wow-agent — no prior agent experience needed.

What is wow-agent, in one paragraph?

It is a coding partner that lives in your terminal. You describe a goal ("fix this bug", "add this feature", "why is this failing?"), and it plans the work, executes it with real tools (reading files, editing code, running tests), verifies the result, and rolls back automatically if verification fails. You can bring any AI model — cloud or local — and your code never leaves your machine unless you send it to a cloud provider yourself.

The 5-minute quick path

# 1. Get it (Linux / macOS / WSL)
git clone https://github.com/uuzjw/wow-agent.git && cd wow-agent
curl -LsSf https://astral.sh/uv/install.sh | sh

# 2. Install dependencies
uv sync

# 3. Launch — the wizard sets up your model
uv run wow

Detailed instructions for every platform (including offline setups): Installation.

Two minutes on the safety model

An agent that runs commands must be constrained. wow-agent ships with four protective layers, all ON by default:

1. Network sandbox — the agent's shell cannot reach the internet
2. Egress guard — git push, scp, curl POST are blocked or confirmed
3. Dangerous-command guard — rm -rf / class commands require typing YES
4. Rollback — every file write is snapshotted; /undo reverts anything

You can read the full story in Safety. For now, know this: a failed task cannot wreck your repo — it rolls itself back.

Your first real task

Read Your First Task — a guided walkthrough showing every screen: the plan, the execution, the verification, and what to do when something goes wrong.

After that

I want to…	Go to
Copy-paste prompts that work	Everyday Cookbook
Master every command	Commands
Understand models & costs	Model Ecosystem
Know how it works inside	Architecture
Answers to common worries	FAQ

How the agent thinks (30-second version)

you give a goal
  → PLANNING   it reads relevant code, proposes a task tree
  → EXECUTING  it works step by step, snapshotting every write
  → VERIFYING  it runs your tests to prove the result
  → DONE       or FAILED → automatic full rollback

That loop is the whole product. Everything else is detail.