Pull core for modelschemas build-time schema and type generation: selection grammar (provider[/endpointGlob][#request|#response]), conditional fetch (ETag/304), the .manifest.json lockfile, atomic file writes, offline build verification, and content-hash integrity checks.

Ships built ESM (.js + .d.ts). Node loads it with no TypeScript loader. Used by @modelschemas/vite and the modelschemas CLI.

bun add @modelschemas/codegen

import { pull, verify, verifyIntegrity, checkUpdates } from '@modelschemas/codegen'

await pull({
  outDir: 'src/modelschemas',
  selections: ['anthropic/v1/messages#request', 'openai/chat/*'],
})

// offline (production builds): files match the manifest
const files = await verify({ outDir: 'src/modelschemas', selections: [...] })

// online: re-fetch each pinned ?version=<contentHash>, re-hash locally
const integrity = await verifyIntegrity({ outDir: 'src/modelschemas', selections: [] })