@bantai-dev/vercel-ai

Vercel AI SDK provider for @bantai-dev/llm. Works with any model supported by the AI SDK (OpenAI, OpenRouter, Anthropic, etc.) and supports structured output via Zod.

Features

• Vercel AI SDK – Uses generateText from the ai package with your chosen model.
• Any model – Pass any LanguageModel (e.g. createOpenAI(), createOpenRouter(), createAnthropic()).
• Zod structured output – Pass outputSchema in the LLM input; uses Output.object({ schema }) under the hood.
• Policy & token quotas – Use with generateText() from @bantai-dev/llm and policies/context from core.

Installation

pnpm add @bantai-dev/vercel-ai @bantai-dev/llm ai zod

For OpenRouter:

pnpm add @openrouter/ai-sdk-provider

Peer dependencies

• zod ^4.3.5

Usage

With OpenRouter

import { vercelAI } from "@bantai-dev/vercel-ai";
import { createOpenRouter } from "@openrouter/ai-sdk-provider";
import { generateText, withLLMContext } from "@bantai-dev/llm";
import { defineContext, definePolicy } from "@bantai-dev/core";
import { z } from "zod";

const openrouter = createOpenRouter({ apiKey: process.env.OPENROUTER_API_KEY });
const provider = vercelAI(openrouter("openai/gpt-4o")); // or e.g. "upstage/solar-pro-3:free"

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

const llmContext = withLLMContext(appContext, { storage });
const policy = definePolicy(llmContext, "My Policy", [/* rules */]);

const result = await generateText({
  provider,
  policies: [policy],
  input: {
    llm: {
      prompt: [
        { role: "system", content: "You are a helpful assistant." },
        { role: "user", content: "Say hello in JSON with a 'message' field." },
      ],
      maxTokensPerRequest: 256,
      outputSchema: z.object({ message: z.string() }),
    },
    tier: "free",
    userId: "user-1",
  },
});

console.log(result.output); // { message: "..." }

With OpenAI (AI SDK)

import { createOpenAI } from "@openai/ai-sdk-provider"; // or from "ai/openai"
import { vercelAI } from "@bantai-dev/vercel-ai";

const openai = createOpenAI({ apiKey: process.env.OPENAI_API_KEY });
const provider = vercelAI(openai("gpt-4o"));

Provider options

Pass through options to the AI SDK generateText call:

const result = await generateText({
  provider,
  policies: [policy],
  input: { llm: { prompt: "Hello" }, /* context */ },
  providerOptions: {
    // e.g. maxTokens, temperature, etc. – depends on AI SDK
  },
});

Prompt formats

• String – Single user message: prompt: "Hello".
• Messages – Array of { role: "user" | "system" | "assistant", content: string } passed to the AI SDK as messages.

Utilities

convertPromptToVercelAIMessages(prompt)

Converts the LLM prompt (string or message array) to the ModelMessage[] format expected by the AI SDK. Used internally by the provider.

convertUIMessagesToPrompt(uiMessages)

Converts Vercel AI SDK UIMessage[] (e.g. from a chat UI) to the prompt format expected by @bantai-dev/llm (array of { role, content }). Useful when wiring UI to generateText with policies.

API

vercelAI(model)

Creates an LLMProvider that delegates to the Vercel AI SDK generateText.

• model – A LanguageModel from the AI SDK (e.g. from createOpenRouter(), createOpenAI()).

Returns an adapter with:

• providerName: "vercel-ai"
• defaultModel: model.modelId or the string model id
• generateText(input, options) – Calls AI SDK generateText with:
  • messages from convertPromptToVercelAIMessages(input.llm.prompt)
  • Optional output: Output.object({ schema: input.llm.outputSchema }) when outputSchema is set
  • Merged providerOptions from the second argument

Environment

• OpenRouter: OPENROUTER_API_KEY
• OpenAI: OPENAI_API_KEY (when using OpenAI via AI SDK)

Related

• @bantai-dev/llm – Context, token quota rules, generateText/streamText.
• Vercel AI SDK – Documentation and providers.

License

MIT