Middleware

wrapProvider returns a provider with middleware applied. Middleware runs in array order.

import { wrapProvider, loggingMiddleware, headerMiddleware, retryMiddleware } from 'vue-llm'

const ai = wrapProvider(createOpenAI({ apiKey }), [
  loggingMiddleware('chat'),
  headerMiddleware({ 'x-team': 'support' }),
  retryMiddleware(),
])

Built-in middleware

Middleware	Description
loggingMiddleware(label?)	Logs requests, finish reasons and errors
headerMiddleware(headers)	Adds providerOptions.headers
retryMiddleware(options?)	Re-sends failed requests with backoff — see Retries
rateLimitMiddleware(options)	Client-side request queue — see Rate Limiting

Writing middleware

import type { Middleware } from 'vue-llm'

const redact: Middleware = {
  beforeRequest(request) {
    return {
      ...request,
      messages: request.messages.map(m =>
        typeof m.content === 'string' ? { ...m, content: m.content.replace(/\d{16}/g, '[card]') } : m),
    }
  },
}

Hook	Signature	Use for
beforeRequest	(request, provider) => request	Rewrite the request
wrapChat	(next, request) => Promise<result>	Call the provider zero or more times (retries, queues, caching)
afterStream	(chunk, request) => chunk | null	Transform or drop stream chunks
onError	(error, request) => Error | void	Replace or observe errors
onFinish	(request, result) => void	Called once the stream is returned

wrapChat composes like an onion: the first middleware in the array is the outermost.