Vercel AI SDK¶

The AI SDK's providers take a baseURL. Create the provider with Privyx's address, and generateText, streamText, and tool calls go through it.

OpenAI¶

privyx proxy --upstream https://api.openai.com

import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";

const openai = createOpenAI({ baseURL: "http://localhost:8000/v1" });

const { text } = await generateText({
  model: openai("gpt-4o-mini"),
  prompt: "Write a short greeting to alice@example.com",
});
console.log(text);

The provider calls OpenAI's Responses API for this model, on /v1/responses, which Privyx masks like Chat Completions.

Streaming:

import { createOpenAI } from "@ai-sdk/openai";
import { streamText } from "ai";

const openai = createOpenAI({ baseURL: "http://localhost:8000/v1" });

const result = streamText({
  model: openai("gpt-4o-mini"),
  prompt: "Write a short greeting to alice@example.com",
});
for await (const delta of result.textStream) {
  process.stdout.write(delta);
}
console.log();

Anthropic¶

privyx proxy --upstream https://api.anthropic.com

import { createAnthropic } from "@ai-sdk/anthropic";
import { generateText } from "ai";

const anthropic = createAnthropic({ baseURL: "http://localhost:8000/v1" });

const { text } = await generateText({
  model: anthropic("claude-opus-5-5"),
  prompt: "Write a short greeting to alice@example.com",
});
console.log(text);

Unlike the Anthropic SDK itself, this provider's baseURL includes /v1.

Without changing code¶

The default openai provider reads OPENAI_BASE_URL:

export OPENAI_BASE_URL=http://localhost:8000/v1

Good to know¶

Run Privyx next to your server code. The base URL above is reached from where the AI SDK runs: your route handlers or server actions, not the browser. In production, use the address of a shared gateway.
Embeddings are not masked. embed and embedMany call /v1/embeddings, which Privyx forwards as sent.