Documentation

Luxe is an agentic coding harness: a terminal program that drives a model through real work in your project, then shows you what it did in a form you can actually review.

If you are here to get started, two pages will do it — Install and Quickstart. Everything else is reference, and you can read it when you hit the thing it describes.

The shape of it

Luxe runs a turn: you say something, the model answers and calls tools, and the loop continues until the work is done or you stop it. Around that sit four ideas worth knowing early.

Roles. Planning, building and reviewing are different jobs, so they are different roles — with different models if you want. See The agent loop.

Guided Review. Rather than a diff, a change arrives as chapters with an argument attached. Comment on a line and the agent picks it up as work. See Guided Review.

Permissions. Permissive by default, because a harness that blocks real work gets its permissions turned off — but with a small set of guardrails that are always on, and a project’s own config can only tighten them. See The security model.

Repo-as-memory. Durable context lives in your repository — task notes, a journal, git itself — not in a conversation that gets compacted away. See Context and compaction.

Getting help from Luxe itself

Luxe ships this manual and can read it. Ask it how to do something and it will look the answer up rather than guess:

how do I point luxe at a local model?

/help lists the key bindings for whatever surface you are on, and every one of them is rebindable — see The interface.