Typed fetch client for modelschemas — live per-endpoint request/response JSON Schemas and model metadata for AI providers (OpenAI, Anthropic, Gemini, xAI Grok, ElevenLabs, OpenRouter, FAL), generated from the service's OpenAPI spec.

Ships built ESM (.js + .d.ts). Node loads it with no TypeScript loader.

bun add @modelschemas/client

import {
  createModelschemasClient,
  listModels,
  getSchema,
  validatePayload,
} from '@modelschemas/client'

const client = createModelschemasClient({
  baseUrl: 'https://modelschemas.com',
})

const models = await listModels({ client, query: { activity: 'chat' } })
const schema = await getSchema({
  client,
  path: { provider: 'anthropic', activity: 'chat', endpointId: 'v1/messages' },
  query: { kind: 'input' },
})

Schema responses are content-addressed (contentHash = ETag = ?version= pin) and carry provenance (upstream sourceUrl, sourceHash, fetchedAt, extractorVersion) so every derivation is independently verifiable.

Docs: modelschemas.com/docs · API spec: /openapi.json · Source: github.com/modelschemas/modelschemas