Developers

Ship an app, not a cloud.

Build local-first apps that run on each user's own machine. The AI, the data, and the accounts are already there. You just ship the app.

Half your backend
ships with every user.

Build local-first AI apps with no backend, no API keys, and no inference bill. Egg turns the browser into one, running the models, the data, and the auth on each user's own device.

Start building Read the docs

import { defineEgglet } from "@egg/sdk";

export default defineEgglet({
  async activate(ctx) {
    // The user's own model. No key, no bill.
    const ai = await ctx.ai.text.chat.create();
    const reply = await ai.prompt("Summarize my week");

    // The user's own data, on their machine.
    await ctx.database.exec(
      "INSERT INTO {{notes}} (body) VALUES (?1)", [reply]
    );
  },
});

The shift

Cloud and client, not cloud or client.

AI is moving to the edge, but it isn't leaving the cloud. The future is hybrid: shared and heavy work in the cloud, personal and instant work on the device. You've always had the cloud half. Egg is the other one.

The economics

Your costs don't scale with your users.

A normal AI app puts every user's inference, data, and identity on your servers, so the bill climbs with every signup. Egg moves that work to the machine the user already owns, and you keep the cloud for what's genuinely shared.

Inference you don't pay for

Routine, per-user prompts run on the user's own model, local or their key. Reach for the cloud only for the heavy lifts.

Less to run

Each user's storage, auth, and sync live on their device. You run servers for what's shared, not for everyone.

No secrets to hold

You don't store API keys or user tokens. Access is brokered on the user's machine, by permission, per app.

Less to leak

The user's private data stays on their device, so it isn't sitting in your database waiting for a breach.

Get started

Zero to a running app in an afternoon.

Scaffold with the SDK and you're calling AI and writing data in a few lines. Or paste our starter prompt into your coding assistant and let it build the first version.