Build on idapt: SDK Quickstart

The @idapt/sdk is a typed client for the whole v1 API: chats, models, Drive, agents, memory, computers. If you can write TypeScript, your workspace is programmable in an afternoon.

Install and authenticate

npm install @idapt/sdk

import { IdaptClient } from "@idapt/sdk";

const client = new IdaptClient({
  apiKey: process.env.IDAPT_API_KEY, // Settings → API Keys
});

Keys are scoped: grant this one exactly the resources your integration touches.

Chat with any model

const reply = await client.chats.send({
  model: "anthropic/claude-sonnet-4.6",
  message: "Three risks in this changelog, one line each: ...",
});
console.log(reply.text);

Swap the model id for anything in the catalog; the call shape stays identical, which is the point of building on a gateway.

Files in, files out

const file = await client.drive.upload({
  path: "Reports/latest.csv",
  content: csvBuffer,
});

const summary = await client.chats.send({
  model: "google/gemini-2.5-flash",
  message: "Summarize the anomalies in this file.",
  attachments: [file.id],
});

Run an agent

const run = await client.agents.run("research-agent", {
  task: "Update /Research/competitors.md with this week's changes",
});

The agent uses its own memory, tools, and permissions: your code delegates work rather than reimplementing it.

Good to know

• The SDK is generated from the same contract as the API, CLI, and MCP surface: one vocabulary everywhere.
• Prefer raw HTTP or another language? The OpenAI-compatible endpoint covers inference, and the API reference documents every route.
• Webhook-style patterns work today via scheduled agents polling state; see automations.

Start with the chat call wired to a real model, then add Drive: the developer hub links the reference, and your first integration is smaller than you think.