Skip to main content

Packaged Install

Use this guide when you want the product-style install flow instead of working from a source checkout.

The packaged path is Composer-first:

  1. install slmcortex
  2. run slmcortex doctor
  3. point it at a folder
  4. compose a runtime
  5. run or export the result

Advanced Factory commands still exist, but they are not the normal first-user path.

For macOS and Linux, the first install path should be Homebrew via alvarolorentedev/homebrew-SLMCortex.

brew tap alvarolorentedev/SLMCortex
brew install slmcortex
slmcortex --help
slmcortex-composer --help

Or install directly from the tap:

brew install alvarolorentedev/SLMCortex/slmcortex

The tap works with Homebrew on macOS and Linuxbrew on Linux and installs the base Composer-first CLI path. Optional runtime backends can be provisioned after install from the CLI itself.

Platform Notes​

The source repo currently ships platform-specific installer scripts:

  • macOS: scripts/installers/install-slmcortex-macos.sh
  • Linux: scripts/installers/install-slmcortex-linux.sh
  • Windows: scripts/installers/install-slmcortex-windows.ps1

The installed package exposes two launchers:

  • slmcortex for the full CLI
  • slmcortex-composer for the guided Composer app entry point

If you are writing docs or onboarding other users, prefer the Homebrew path on macOS and Linux. Keep the repo installer scripts and source checkout path as the lower-level fallback.

Workspace Contract​

The packaged app keeps state outside the repository you are composing.

Typical workspace folders:

  • state/ for local state and copied demo repos
  • packages/ for imported or authored SLM packages
  • runtimes/ for emitted runtime bundles
  • exports/ for handoff descriptors
  • logs/ for compose and diagnostics logs
  • diagnostics/ for doctor exports and support bundles

Inspect the resolved workspace at any time:

slmcortex doctor
slmcortex doctor --workspace /tmp/slmcortex-app
slmcortex doctor --export-support-bundle
slmcortex-composer --help

Default Product Flow​

Start with readiness checks:

slmcortex doctor
slmcortex compose-folder --help

Compose a runtime from one folder:

slmcortex compose-folder \
  --workspace /tmp/slmcortex-app \
  --folder /path/to/repo \
  --task "Create a FastAPI endpoint with request validation" \
  --export-descriptor /tmp/slmcortex-app/exports/repo.json

Or launch the guided Composer flow directly:

slmcortex-composer \
  --workspace /tmp/slmcortex-app \
  --folder /path/to/repo \
  --task "Create a FastAPI endpoint with request validation" \
  --outcome export_bundle \
  --export-logs

Optional Runtime Provisioning​

The base install is meant to stay usable even before real-model dependencies are installed.

When you want actual local inference:

slmcortex provision-backend --backend mlx --dry-run
slmcortex provision-backend --backend gguf
  • MLX is the Apple Silicon path.
  • GGUF covers Linux, Windows, macOS Intel, and explicit GGUF use.

What This Flow Is Good For​

Use the packaged flow when you want:

  • a product-style launcher instead of repo-relative commands
  • a separate app workspace for packages and runtimes
  • one-step folder scan, routing, composition, and export

Use the source quickstart when you want:

  • the fastest developer validation path
  • direct access to demo scripts and checked-in fixtures
  • the full Factory and runtime surface from a checkout