Setup

Install the provider adapter beside Anvia core:

pnpm add @anvia/core @anvia/openai

Set OPENAI_API_KEY in the server environment, then construct the client in server-only code:

import { OpenAIClient } from '@anvia/openai'

export const openai = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY,
})

Do not place the client or API key in a browser bundle. Browser applications should call an application route that owns the model request.

Client options

OpenAIClient accepts the following verified configuration:

Option	Use
apiKey	Authenticate a client created by the adapter.
baseUrl	Target an OpenAI-compatible endpoint.
headers	Add headers required by the configured endpoint.
completionApi	Select "responses" or "chat".
client	Reuse an already-created OpenAI SDK client.

The constructor requires either apiKey or client. Keep a preconfigured client useful when transport, organization-wide SDK configuration, or application dependency injection belongs elsewhere.

import OpenAI from 'openai'
import { OpenAIClient } from '@anvia/openai'

const sdk = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
})

export const openai = new OpenAIClient({ client: sdk })

Custom endpoints

Setting baseUrl makes the Chat adapter the default. Compatible endpoints often share request shapes without supporting the same tools, schemas, reasoning fields, streaming chunks, or media behavior.

const compatible = new OpenAIClient({
  apiKey: process.env.COMPATIBLE_API_KEY,
  baseUrl: 'https://provider.example.com/v1',
  completionApi: 'chat',
})

Use the dedicated Compatible APIs guide and smoke test every required workflow. The current documented @anvia/openai public surface does not provide a dedicated Azure OpenAI client, so this guide does not prescribe unverified Azure configuration.

Export models, not credentials

A small model module keeps provider details out of agents and business logic:

export const supportModel = openai.completionModel('gpt-5')
export const knowledgeEmbeddings = openai.embeddingModel(
  'text-embedding-3-small',
)

Model-name types provide autocomplete for known IDs while still allowing custom strings. Validate a configured model ID at startup or with a deployment smoke test rather than assuming autocomplete proves availability.