@karpo.dev/sdk-core

The transport-agnostic Karpo client: the local SQLite replica, the query and watch layer, the outbox, the sync loop, and the auth session. Everything environmental — opening a database, fetch, clocks, id generation — arrives through a Platform seam, which is what lets the same core run in a browser worker, in Node, and in tests without behaving differently.

npm install @karpo.dev/sdk-core

Applications normally install a runtime package instead — @karpo.dev/sdk-web in the browser — and get this one as a dependency.

Entry points

| Import | What it gives you | | --- | --- | | @karpo.dev/sdk-core | createCoreClient, the runtime interfaces, the query layer, sync and outbox types | | @karpo.dev/sdk-core/node | createNodeClient / createNodePlatform over Node's built-in node:sqlite | | @karpo.dev/sdk-core/testing | The in-memory fixture and transport stub the suites drive |

Usage

import { createNodeClient } from "@karpo.dev/sdk-core/node";
import schema from "./karpo/generated/schema.json";

const karpo = await createNodeClient({
  schema,
  user: "u01",
  dataDir: ".karpo/replica",
  target: "dev",
});

await karpo.runtime.insert("workspaces", "todos", {
  containerId: "ws-demo",
  title: "Write the README",
});

// Re-runs after any committed transaction touching the table.
const stop = karpo.runtime.watchTables(["workspaces.todos"], () => {
  void karpo.runtime.selectSql("select * from todos", []).then(render);
});

Reads are local and synchronous-fast; writes commit locally, enter the outbox, and push in the background, so the client works offline and converges when the connection returns. selectSql rejects anything that is not a single SELECT — it is the seam the typed query layer compiles onto.

Documentation

• Queries: https://karpo.dev/queries
• Sync: https://karpo.dev/sync
• Transactions: https://karpo.dev/transactions
• Source: https://github.com/Chirich-GmbH/karpo-core (packages/sdk-core)