Package one capability
Wrap a focused LoRA adapter as a reusable SLM package with provenance, checksums, and routing metadata.
Package focused coding capabilities. Compose a local runtime. Validate before trust.
SLMCortex turns LoRA adapters into reusable SLM packages, composes them into a deterministic runtime bundle, and uses the same runtime for dry-run validation, local inference, a compatibility server, and bounded repository work.
public pathdry run first
$ python -m slmcortex doctor $ brew install alvarolorentedev/SLMCortex/slmcortex $ slmcortex doctor inspect workspace check backends compose-folder --help source checkout fallback: python scripts/run_slmcortex_demo.py
The package is the unit of distribution.The runtime bundle is the unit of deployment.The runtime is the unit of execution.
One explicit path from small brain to local engine
Visitors should be able to understand the product without reading the source. The core flow is package, compose, validate, then run.
Compose one runtime
Combine validated SLM packages into a deterministic runtime bundle without mutating the source packages.
Validate before trust
Check package and runtime structure before inference, serving, or agent behavior touches a local repository.
Run the same core everywhere
Use Runtime Core for dry-run routing, local inference, the compatibility server, and bounded agent runs.
Why teams reach for this instead of another hosted handler
The pitch is not "bigger models." The pitch is clearer boundaries: local artifacts, explicit runtime assembly, and dry-run validation before you spend time on real backend setup.
Common friction
Paid LLM services can turn every coding workflow into a metered remote dependency.
SLMCortex path
SLMCortex keeps focused coding capabilities local, packaged, and inspectable before they run.
Common friction
One large general model is often asked to do jobs that are narrow, repetitive, and expensive to validate.
SLMCortex path
SLMCortex packages smaller task-shaped capabilities and composes them into one runtime bundle.
Common friction
Tooling teams often need proof that routing and control flow work before they take on real backend complexity.
SLMCortex path
Start with dry-run validation, then move to local inference only when your backend and model setup are ready.
Proof you can inspect today
Start with dry-run validation. Move to real model-backed inference only when your local backend and model setup are ready.
Homebrew install
macOS and Linux users can start with the Homebrew tap instead of cloning the source repo first.
No-model demo
Checks package, compose, validate, infer, and agent flow without downloading weights.
Composer-first path
Use doctor and compose-folder when you want the product-style install and folder-to-runtime flow.
Backend choices
MLX is used on Apple Silicon; GGUF covers Linux, Windows, macOS Intel, and explicit GGUF use.
Bounded agent
The current agent is local and single-run, with writes controlled by flags rather than hidden background behavior.
Honest v0.1 boundaries
SLMCortex is easiest to evaluate because its limits are visible. The current release is a narrow local path, not a broad production-agent platform.
- Local, single-run execution only.
- Bounded tool loop, not a full IDE agent.
- Real inference requires local backend and model setup.
- The compatibility server is intentionally minimal and non-streaming.
Choose the next technical path
Use the docs to verify the demo, understand the packaged flow, or inspect the runtime architecture.
Quickstart
Start with the smallest end-to-end path and verify the runtime on your machine.
Packaged Install
Use the Composer-first install flow with doctor, compose-folder, and the external workspace contract.
Command Reference
See the public CLI surface for packaging, composition, inference, serving, and agent work.
Architecture
Understand Factory, Composer, Runtime Core, and Agent Runtime without reading the source first.