@anvia/openai: Examples

Minimal agent

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

const client = new OpenAIClient({ apiKey: process.env.OPENAI_API_KEY });
const agent = new AgentBuilder("support", client.completionModel("gpt-5"))
  .instructions("Answer support questions clearly.")
  .build();

const response = await agent.prompt("Draft a short support reply.").send();
console.log(response.output);

Product-shaped model boundary

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

export function createSupportModel(): CompletionModel {
  const client = new OpenAIClient({ apiKey: process.env.OPENAI_API_KEY });
  return client.completionModel("gpt-5");
}

export function createFallbackModel(): CompletionModel {
  const client = new OpenAIClient({ apiKey: process.env.OPENAI_API_KEY });
  return client.completionModel("gpt-5");
}

The application can now choose a model before building the agent while keeping the agent factory provider-neutral.

Harness shape

import { describe, expect, it } from "vitest";

describe("support model boundary", () => {
  it("creates a completion model", () => {
    const model = createSupportModel();

    expect(model).toHaveProperty("completion");
    expect(model).toHaveProperty("streamCompletion");
  });
});

Use live provider tests sparingly and gate them behind environment variables. Unit tests should usually check the model boundary and mock the model contract.