Vercel AI SDK Integration

Scavio ships as a set of ready-made tools for the Vercel AI SDK. Import the tools from @scavio/ai-sdk and pass them to generateText or streamText to give any AI SDK agent real-time search across Google, YouTube, Reddit, Amazon, Walmart, TikTok, and Instagram — plus scavio_extract, which reads any URL. A cost-effective Tavily and SerpAPI alternative, with one package, one API key, and no custom HTTP code.

Introduction

@scavio/ai-sdk exposes Scavio endpoints as standard AI SDK tool() objects with typed Zod input schemas. Pass scavioTools() for the full set, or import individual factories to expose a lean tool list to the model. You need Node.js 18 or later and a Scavio API key from dashboard.scavio.dev.

The scavio JS SDK is bundled and called under the hood — it handles auth, rate limiting, and request formatting, and returns each result to the model as the raw Scavio JSON response.

Step-by-Step Integration Guide

Step 1: Install

npm install @scavio/ai-sdk ai zod

ai and zod are peer dependencies; the scavio JS SDK is bundled and called under the hood. Requires Node.js 18 or later.

Step 2: Set your API key

Get a key at dashboard.scavio.dev (free credits, no card), then set it as an environment variable:

export SCAVIO_API_KEY=sk_live_your_key

Each tool factory reads SCAVIO_API_KEY from the environment. You can also pass it explicitly: scavioSearch({ apiKey: "sk_live_..." }).

Step 3: Basic usage

import { generateText, stepCountIs } from "ai";
import { openai } from "@ai-sdk/openai";
import { scavioTools } from "@scavio/ai-sdk";

const { text } = await generateText({
  model: openai("gpt-5.5"),
  tools: scavioTools(),
  stopWhen: stepCountIs(3),
  prompt: "Find the official GitHub repo of the Agno framework and summarize it",
});

console.log(text);

Available Tools

scavioTools() returns all 13 tools keyed by name, ready to spread into the tools option. Or import factories individually.

Every tool is also exported as its own factory, so you can expose a lean tool list to the model:

import { generateText, stepCountIs } from "ai";
import { openai } from "@ai-sdk/openai";
import { scavioSearch, scavioAmazonSearch } from "@scavio/ai-sdk";

const { text } = await generateText({
  model: openai("gpt-5.5"),
  tools: {
    scavio_search: scavioSearch({ maxResults: 5 }),
    scavio_amazon_search: scavioAmazonSearch(),
  },
  stopWhen: stepCountIs(3),
  prompt: "Compare prices for a mechanical keyboard on Amazon",
});

Each factory accepts { apiKey?, maxResults?, ...ScavioConfig }. maxResults trims long results arrays before they reach the model (defaults to 10), keeping token usage down. It does not apply to scavio_extract, which returns page content rather than a result list.

Search, then read

scavio_extract is what makes the curated set self-sufficient: the agent searches with one tool, then reads any URL the result set points at — including pages on platforms this package does not wrap.

import { generateText, stepCountIs } from "ai";
import { openai } from "@ai-sdk/openai";
import { scavioSearch, scavioExtract } from "@scavio/ai-sdk";

const { text } = await generateText({
  model: openai("gpt-5.5"),
  tools: {
    scavio_search: scavioSearch({ maxResults: 5 }),
    scavio_extract: scavioExtract(),
  },
  stopWhen: stepCountIs(4),
  prompt: "Find the Bun 2.0 release notes, open the page, and list the breaking changes.",
});

format is markdown (default), text, or html. mode sets the fetch tier and the price: normal (default) is a plain fetch, advanced renders the page in a browser for JS-built sites, and ultra routes through residential proxies for the hardest bot walls. Only a successful read is billed.

Advanced Example

Stream a multi-step research run: the model chains web and Reddit searches on its own, then narrates its findings token by token.

import { streamText, stepCountIs } from "ai";
import { openai } from "@ai-sdk/openai";
import { scavioTools } from "@scavio/ai-sdk";

const result = streamText({
  model: openai("gpt-5.5"),
  tools: scavioTools(),
  stopWhen: stepCountIs(5),
  system: "You are a research assistant. Use Scavio for fresh web data and cite sources.",
  prompt: "Research Tavily alternatives: check the web, then see what developers on Reddit say.",
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

The agent picks scavio_search and scavio_reddit_search as it goes, up to the stepCountIs(5) budget — you never wire up that routing yourself.

How it works

Each tool is a standard AI SDK tool() with a typed Zod input schema, so the model gets accurate argument hints and the SDK validates calls before they run. Calls go through the scavio JS SDK, which handles auth, rate limiting, and request formatting. Tool results are returned to the model as the raw Scavio JSON response.

Credit costs

Costs are not uniform, and each tool states its own in the description the model sees. Most calls cost 1 credit, including Google search, Reddit search, both Amazon tools, Walmart and TikTok. YouTube search costs 2 and scavio_youtube_transcript costs 8. Instagram user search costs 10. scavio_extract is priced by the mode you ask for — 1 on normal and advanced, 2 on ultra — not per call. See the rate limits reference for plan limits and the errors reference for retry guidance.

Benefits of Scavio + Vercel AI SDK

• Drop-in tools: spread scavioTools() into generateText or streamText and go.
• Typed Zod schemas: accurate argument hints for the model and validated calls before they run.
• Curated, not exhaustive: 13 tools over seven platforms plus Extract, short enough for a model to route over reliably — with the MCP server and the JS SDK there when you need all 31 platforms.
• Search, then read: scavio_extract turns any URL a search returns into clean Markdown, with no headless browser to run.
• Cost-effective: most calls cost a single credit — a Tavily and SerpAPI alternative with broader platform coverage.

Next Steps

• Scavio API quickstart — keys, credits, and your first request
• Google Search API reference — the endpoint behind scavio_search
• Extract API reference — the endpoint behind scavio_extract
• MCP Integration — every Scavio endpoint as a tool
• JavaScript SDK reference — the client these tools call, with every platform namespace
• @scavio/ai-sdk on npm
• Vercel AI SDK documentation
• scavio JS SDK on npm