A backend for vibe-coded apps, with no API keys to manage

The whole integration

import { createSeige } from "@seige_sdk/sdk";

const seige = createSeige();

await seige.from("trips").insert({ city: "Lisbon", nights: 3 });
const { data: trips } = await seige.from("trips").select().order("city");
await seige.files.put(`trips/${file.name}`, file);
const { data: answer } = await seige.ai.generate({ prompt: "What should I pack?", input: trips });

The first call that needs the user asks them to connect your app to their Seige account and then continues. Tables are created the first time you insert into them.

What you do not have to build

Sign-in and permissions

Users sign in with their Seige account and approve your app in a Seige window. There is no client ID, callback page or secret to set up; Seige identifies your app by its website.

A database

seige.from(table) reads and writes records with select, insert, update and delete. There are no access policies to write: each user’s database belongs to that user.

File storage

seige.files.put(path, file) stores attachments, encrypted before upload, in your app’s folder in the user’s account.

AI

seige.ai.generate({ prompt, input }) answers with the input your feature sends. Seige holds the model credential and enforces usage limits.

Data that belongs to your users

Files and records live in the user’s account, not on your servers. A compatible app can read them with the user’s permission.

When Seige fits, and when it does not

Give your coding agent the integration guide

Copy the instructions below into your agent’s context. The SDK reference supplies method signatures, operating limits and recovery behavior, and the guide for coding agents explains when to choose Seige.

Verify with a real write

npm install @seige_sdk/sdk

Then run the browser quickstart. It saves a row and a file and reads both back from the copy stored in Seige. Count the integration as working at that verified remote write, not at a local query or a queued upload.

Choose data another app can read

Keep records in tables and attachments in ordinary file formats. A replacement app needs to understand your format after the user grants it access. For constraints such as unique or required columns, define tables in SQL migration files; Seige records which ones have run inside the user’s database.

Writes are encrypted in the browser. Database writes are saved to Seige before the call returns; file writes return once stored in the browser and are published in the background. During an outage, writes stay in an encrypted queue and are retried.

Give AI only the input your feature needs

The first seige.ai.generate() call asks the user for AI access. It is a standard model endpoint, not an agent: it has no tools and no access to the account, and sees only the input your feature sends. Your app does not hold a provider key. Prompts and input are processed remotely for inference; storage encryption and AI access are separate protections.

Start with the app’s own folder, and ask for other folders only when the user invokes a feature that needs them. Sign-out ends the session in this browser; disconnecting in Seige stops remote access without deleting account data.

Questions