Quickstart

Get from nothing to a real answer and a real code review in about five minutes.

You need Node.js 24+ and an API key from one AI provider (or a local Ollama — see Local & free models if you don’t want to pay for one).

1. Install

npm install -g gaunt-sloth

This gives you the gth command (with gsloth and gaunt-sloth as aliases — use whichever you like; this guide uses gth).

2. Set your provider key

Export the key for the provider you have. For Anthropic:

export ANTHROPIC_API_KEY="sk-ant-..."

Other providers read their own variable — GEMINI_API_KEY, OPENAI_API_KEY, GROQ_API_KEY, OPENROUTER_API_KEY, and so on. The full list is in Providers.

3. Initialize the project

From your project’s root:

gth init anthropic

This writes a .gsloth.config.json you can commit and tune later. The provider name is one of vertexai, anthropic, groq, deepseek, openai, google-genai, xai, openrouter, huggingface, ollama.

Prefer to be walked through it? Run gth init with no argument — it detects which providers have a key set, lets you pick a provider and model, and offers to store the config for this project or globally. Interactive init can also target the global config or a named profile with -g / -i <name> — see Creating a profile.

4. Ask your first question

gth ask "what does this project do?" -f README.md

ask is a one-shot question. -f adds files as context (repeat it for more), and you can pipe input in:

cat error.log | gth ask "what is causing these errors?"

5. Review your changes

Reviews are stateless — nothing carries over between runs — so a review can’t be argued down, and a failing one exits non-zero, which is what makes it a CI gate.

git --no-pager diff | gth review

Working on a pull request? gth pr 42 reviews PR #42 and pulls its linked issue in as the requirements to check against.

Where to go next

• Work interactively — Interactive sessions: bare gth opens a coding session in your project; chat mode, slash commands, the TUI.
• Do a specific job — Guides & Recipes: review PRs in CI, code against your own project rules, run a local model, script it non-interactively.
• Tune the setup — Configuration: providers, tools, MCP servers, identity profiles.
• Every command and flag — Commands.