@bantai-dev/openai

OpenAI provider for @bantai-dev/llm. Uses the OpenAI Responses API with optional Zod-based structured output.

Features

• OpenAI Responses API – Uses openai.responses.create() for chat and structured output.
• Zod structured output – Pass outputSchema in the LLM input; responses are validated and typed.
• Policy & token quotas – Use with generateText() from @bantai-dev/llm and policies/context from core.

Installation

pnpm add @bantai-dev/openai @bantai-dev/llm openai zod

Peer dependencies

• zod ^4.3.5

Usage

Basic

import { openai } from "@bantai-dev/openai";
import { generateText, withLLMContext } from "@bantai-dev/llm";
import { defineContext, definePolicy } from "@bantai-dev/core";
import { z } from "zod";

const appContext = defineContext(
  z.object({
    userId: z.string().optional(),
    tier: z.enum(["free", "premium"]),
  })
);

const llmContext = withLLMContext(appContext, { storage }); // storage: see @bantai-dev/llm README
const policy = definePolicy(llmContext, "My Policy", [/* rules */]);

const provider = openai("gpt-4o");

const result = await generateText({
  provider,
  policies: [policy],
  input: {
    llm: {
      prompt: [{ role: "user", content: "Hello" }],
      maxTokensPerRequest: 512,
      outputSchema: z.object({ answer: z.string() }),
    },
    tier: "free",
    userId: "user-1",
  },
});

console.log(result.output); // { answer: "..." }
console.log(result.usage);  // { inputTokens, outputTokens, totalTokens }

With client options

import { openai } from "@bantai-dev/openai";

const provider = openai("gpt-4o", {
  providerOptions: {
    apiKey: process.env.OPENAI_API_KEY,
    // other OpenAI ClientOptions
  },
});

Prompt formats

• String – Converted to a single user message: prompt: "Hello".
• Messages – Array of { role: "user" | "system" | "assistant", content: string } passed to the Responses API.

API

openai(model, options?)

Creates an LLMProvider for the OpenAI Responses API.

• model – OpenAI chat model id (e.g. "gpt-4o", "gpt-4o-mini"). See OpenAI models.
• options
  • providerOptions – Optional OpenAI ClientOptions (e.g. apiKey, baseURL).

Returns an adapter with:

• providerName: "openai"
• defaultModel: the given model
• generateText(input, options) – Calls openai.responses.create() with:
  • input from convertPromptToOpenAIMessages(input.llm.prompt)
  • Optional text: { format: zodTextFormat(outputSchema, "output") } when input.llm.outputSchema is set

convertPromptToOpenAIMessages(prompt)

Utility to convert the LLM prompt (string or message array) to the format expected by openai.responses.create() input. Re-exported for use in custom wrappers if needed.

Environment

• OPENAI_API_KEY – Set for the OpenAI client (or pass via providerOptions.apiKey).

Related

• @bantai-dev/llm – Context, token quota rules, generateText/streamText.
• OpenAI Node SDK – Underlying client.

License

MIT