Skip to content

Configure projects and models

Edit intloom.yaml while the Service is stopped, preserving installed looms entries. The generated minimum is:

localStorage: file
useMcpAgent: true
looms: []

Choose file or sqlite before saving business data; changing a populated backend has no automatic migration.

Configure service models

llms:
  default:
    provider: openai-compatible
    model: your-default-model
    secret: ENV.MODEL_API_KEY
    baseURL: https://your-provider.example/v1
  working:
    provider: anthropic
    model: your-working-model
    secret: ENV.WORKING_API_KEY
    parameters:
      maxOutputTokens: 8192

Providers are anthropic and openai-compatible. Steps select thinking, working, or coding; omitted roles fall back to default. Each supplied role is a complete model configuration.

ENV.NAME reads process.env first. Only when the variable is undefined does it read the selected project root's regular .env file. For example, secret: ENV.ZEN_API_KEY can use ZEN_API_KEY=your-key in that project's .env. No parent directory, caller working directory, .env.local or symlink file is loaded, and file values never enter global process.env. Quotes, comments and multiline values follow Node's dotenv syntax; variable expansion and decryption are not performed.

Empty or still-encrypted values are rejected, including defined startup values even if .env contains a valid value. DENV.NAME requires an externally decrypted startup environment value. Restart the Service after editing .env; readiness and Agent assembly share the same lazy file snapshot.

intloom stop
intloom start
intloom doctor --execution cli
intloom config show --json

A running Service retains its startup configuration and environment. For revision-checked updates, read the revision from config show and pass it to config set --file FILE --revision HASH. The file is a complete replacement configuration.

See all configuration fields and Service operations.