Global config entry point: ~/.lume/lume.yaml
# Model config
models:
default: openai/gpt-4o
# Workspaces
workspaces:
my-project:
path: ~/projects/my-project
context: .context/
# MCP servers
mcp:
my-server:
command: npx
args: ["-y", "my-mcp-server"]models
default sets the default model as provider/model. Models connect through an OpenAI-compatible API and can be switched per conversation via the picker below the input box.
workspaces
Each workspace binds one local directory:
path: the project path.context: project context directory (relative), where the agent reads structure and conventions.
mcp
Mount MCP servers over stdio: command starts the server, args are its arguments. Their tools appear automatically in the agent’s tool list.
Data Directory
Under ~/.lume/, memories, conversations and configs live in separate plain-text directories — readable and backup-able at any time.