@zap-tools/sdk

Server & client SDK for ZapTools — create a streaming AI chat backend with tool calling, powered by Vercel AI SDK and OpenRouter.

Install

npm install @zap-tools/sdk zod

Server Usage

// app/api/chat/route.ts
import { createZapServer } from "@zap-tools/sdk";
import { tool } from "ai";
import { z } from "zod";

const zap = createZapServer({
  apiKey: process.env.OPENROUTER_API_KEY!,
  systemPrompt: "You are a helpful assistant.",
  tools: {
    getUserEmail: tool({
      description: "Get the current user email",
      parameters: z.object({}),
      execute: async () => "[email protected]",
    }),
  },
});

export async function POST(req: Request) {
  return zap.handleRequest(req);
}

Client Usage

import { useZapChat } from "@zap-tools/sdk/client";

const chat = useZapChat({ endpoint: "/api/chat" });

The /client subpath is tree-shaken from server code — no server deps in your client bundle.

Exports

| Subpath | Description | |---------|-------------| | @zap-tools/sdk | Server — createZapServer, types, ZapTool | | @zap-tools/sdk/client | Client — useZapChat hook, ZapChatState, ZapMessage |

Server Config

| Field | Type | Default | Description | |-------|------|---------|-------------| | apiKey | string | — | Your OpenRouter API key (server-only) | | model | string | openai/gpt-oss-120b:free | Any OpenRouter model ID | | systemPrompt | string | — | System prompt for the AI | | tools | Record<string, ZapTool> | {} | Tools the AI can call server-side | | maxSteps | number | 5 | Max tool-calling loops per request |

License

MIT