Luminary

Turn any codebase into a visual architecture map, then trace the path a feature takes through it.

Luminary is for the first week on an unfamiliar repository: the point where you can read any single file but cannot yet say how the pieces fit together.

What it does

System and container maps. Luminary parses the repository and draws what it finds — the modules, the services they belong to, the external endpoints they call, and the edges between them. Every edge comes from real evidence in the source, not a guess from folder names.

TraceFlow. Describe a task in plain words — "the login page is broken" — and Luminary matches it to a workflow in the repository and draws the path that flow takes through the files. When your description is ambiguous, it asks a follow-up rather than picking for you. When nothing matches, it says so instead of forcing a best guess.

Getting started

Open a folder in VS Code.
Run Luminary: Index Workspace from the Command Palette.
Choose Current Workspace, or GitHub URL to point at a repository you have not cloned.

Indexing runs entirely on your machine. Nothing is uploaded.

Languages

Language	Analysis
TypeScript, JavaScript	Full AST parse: symbols, imports, routes, API calls, config reads
Python	Imports, symbols, and route decorators
Go	Functions, types, and imports

Other text and config files are still indexed so the shape of the repository stays visible.

AI features

Two features use a language model: TraceFlow and Enhance verbs with AI on the Map view. Both are optional — the map itself is built entirely from static analysis and needs no key and no network.

Two ways to turn them on:

Use your own API key — run Luminary: Set API Key and paste an OpenAI or Anthropic key. Luminary talks to that provider directly; usage is billed to you, and nothing passes through us.
Use an access code — if you were given one, run Luminary: Enter Access Code. Nothing else to set up.

Either one is stored in your operating system’s keychain through VS Code’s SecretStorage — never in a settings file, and never synced to another machine.

What is sent. When you use an AI feature, Luminary sends the names of workflows, modules and exported functions, the file paths they live in, and the text you typed. It does not send the contents of your files. To switch this off entirely, set luminary.ai.enabled to false; no request will leave your machine.

Settings

Setting	Default	What it does
`luminary.maxFiles`	`2000`	Files scanned per run. Lower it if indexing is slow on your machine. Luminary tells you when a repository was too large to scan completely.
`luminary.ai.enabled`	`true`	Turn all AI features off.

Commands

Command
`Luminary: Index Workspace`	Build the map
`Luminary: Set API Key`	Use your own OpenAI or Anthropic key
`Luminary: Enter Access Code`	Use an access code, if you were given one
`Luminary: Clear Credentials`	Remove whatever is stored

Requirements

Git must be installed to use the GitHub URL source. Indexing the open workspace does not need it.

Private repositories are cloned using the GitHub account you are already signed in to in VS Code. You are only asked to sign in if a clone is actually refused.

Luminary — Map for your Codebase

Luminary Dev