@anvia/openai

Connect Anvia agents to OpenAI or an OpenAI-compatible endpoint. Use one client for chat, typed decisions, embeddings, images, speech, and transcription.

For Azure OpenAI and Azure AI Foundry, use @anvia/azure for Azure endpoint and authentication configuration.

Install

pnpm add @anvia/openai @anvia/core

Quickstart

Set OPENAI_API_KEY before running this example.

import { Agent } from "@anvia/core";
import { OpenAIClient } from "@anvia/openai";

const client = new OpenAIClient({
  apiKey: process.env.OPENAI_API_KEY!,
});
const agent = new Agent({
  id: "assistant",
  model: client.completionModel({ modelId: "gpt-5.6", api: "responses" }),
  instructions: "Answer clearly and concisely.",
});

const result = await agent.generate({
  prompt: "What should I check before launching a new product?",
});

if (result.type === "response") {
  console.log(result.output);
}

Typed decisions

import { OpenAIClient, GPT_6_LUNA } from "@anvia/openai";
import { check, decide } from "@anvia/core/decision";

const model = new OpenAIClient({ apiKey: process.env.OPENAI_API_KEY! }).decisionModel({
  modelId: GPT_6_LUNA,
});
const result = await decide({
  model,
  state: { message: "Please refund my duplicate charge." },
  questions: { refund: check({ instructions: "Is a refund requested?" }) },
});
console.log(result.answers.refund.probability);

Uses the dedicated Decisions API with OpenAI SDK 7.30.0+. Text and structured JSON are supported; structured state is serialized into text. Refusals throw DecisionRefusalError from Core. See the usage guide for mapping, limits, batches, and refusal handling.

What you can build

• Chat Completions and Responses APIs, streaming, tools, and structured output.
• Typed reasoning controls for supported models.
• Typed classification, multi-label checks, and rubric scoring through decisionModel() and decide().
• Embeddings, image generation, speech generation, transcription, and model listing.
• Custom endpoints through baseUrl.

Model handles default to Chat Completions. The example selects Responses explicitly. Bun users can install the same packages with bun add.

Learn more

• Usage guide
• Anvia overview
• Contributing

Compatible model metadata

listModels() preserves advertised supported_parameters as supportedParameters when it is an array of strings, and object-valued reasoning as provider-specific metadata. Validate reasoning fields before using them; their schema varies by service. Object-valued pricing is preserved the same way; validate its fields before using them.