plgg-kit

plgg-kit is the family's vendored dependency surface — the LLM provider abstractions a plgg project (notably plgg-foundry) builds on. It is not an API to learn in depth; it is the seam where external AI vendors enter the plgg world as plain data.

LLM providers

A provider is a tagged Box carrying a model + optional API key — pure config, not a live client:

Constructor	Type
`openai(model)` / `openai({ model, apiKey? })`	`OpenAI = Box<"OpenAI", Config>`
`anthropic(model)` / `anthropic({ model, apiKey? })`	`Anthropic = Box<"Anthropic", Config>`
`google(model)` / `google({ model, apiKey? })`	`Google = Box<"Google", Config>`

import { openai, anthropic } from "plgg-kit";

const provider = openai("gpt-5.1");
const alt = anthropic({ model: "claude-opus-4-8", apiKey });

The optional apiKey is lifted to an Option (never an undefined literal), and each provider has its as* decoder and *$ matcher, so a provider can be validated and folded by name like any other plgg value.

Why a project depends on it

Keeping the vendor abstractions in one package means the rest of the family stays vendor-neutral: plgg-foundry takes a provider value and never names a vendor SDK directly, so swapping LLM backends is a one-line change. This is the same boundary discipline the HTTP packages apply to runtimes — external concerns confined to a single seam.