About ModelRig

Buying advice you can audit.

Tell us what you want to run. We’ll tell you what to buy.

Why this exists

Most local-AI hardware tools start with a computer you already own. ModelRig starts with the work: model or quality target, context, concurrent agents, budget, operating constraints, and remote operation.

Why there is no recommendation LLM

A purchase decision should not change because a language model sampled a different adjective. The TypeScript engine is deterministic, unit-tested, framework-independent, and shared by the web app and CLI. No OpenAI key, account, database, telemetry, or proprietary API is required.

Open-source boundaries

ModelRig is released under the MIT License. Catalog contributors must attach provenance and uncertainty. Benchmarks must identify the exact—or explicitly comparable—configuration. Unknown is an acceptable answer.

Security model

Generated deployments bind inference services to loopback, expect authentication where the runtime supports it, and use a private Tailscale network plus SSH tunneling. Scripts are inspectable and do not silently execute privileged package installation. No generated setup opens a router port.

Contribute

Read the repository’s CONTRIBUTING, SECURITY, catalog examples, validation scripts, and architecture decision record. Catalog changes should be small, sourced, and independently reviewable.