Getting Started

Install

Grab the latest build from the download page:

• Windows: download the .exe installer and follow the wizard.
• macOS: pick the Apple Silicon or Intel .dmg, open it and drag Lume into Applications.

Building from source? Clone the repo and run bun install && bun build:desktop. You need Node.js ≥ 20, Bun ≥ 1.0, and Rust stable for the desktop native modules.

First Launch

Lume is a local app: once running, everything it knows lives under ~/.lume/ on your machine — memories, conversations, project context, skill configs — all plain files you can read and edit directly.

Configure a Model

The global config entry point is ~/.lume/lume.yaml:

models:
  default: openai/gpt-4o

Lume connects to mainstream models (OpenAI, Anthropic, Gemini, DeepSeek, GLM, Qwen, Doubao, Moonshot…) through an OpenAI-compatible API; you can also switch models from the picker below the chat input.

Create a Workspace

Attach a project directory so the agent can read code and run commands:

workspaces:
  my-project:
    path: ~/projects/my-project
    context: .context/

Start Using It

Open a new thread and describe what you want done. The main thread understands the task, calls tools, and hands subtasks to the best-fit persona when needed; you can interrupt, add context or take over at any time.