@flowget/engine

The workflow execution substrate for Flowget — a Temporal-backed graph walker with a node registry, decorators, lifecycle events, human-in-the-loop signal primitives, and graph validators. It is consumed by the Flowget worker and CLI, and re-exports the node-authoring helpers so node authors need only a single import.

Install

npm i @flowget/engine

Usage

Author a node and build a registry the walker can resolve against:

import { z } from "zod";
import { buildRegistry, defineNode } from "@flowget/engine";

const httpFetch = defineNode({
  id: "http_fetch",
  category: "data",
  input: { url: z.string().url() },
  output: { status: z.number(), body: z.unknown() },
  async run({ input }) {
    const res = await fetch(input.url);
    return { status: res.status, body: await res.json() };
  },
});

const registry = buildRegistry([httpFetch]);

input.url is typed string inside run automatically — the engine never imports a specific validator library, so any object with a .parse() method works (Zod, Valibot, custom).

Production HTTP nodes must validate inputs (URL allow-list, https-only, block private/loopback IPs, cap response size, set a timeout) to prevent SSRF and DoS. The engine ships assertSafeUrl and redactUrlForLogging to help.

⚠️ 0.x is unstable

The surface is unstable until v1 — breaking changes ship as minor bumps. Pin exact versions in consumers ("@flowget/engine": "0.1.0", not "^0.1.0").

Links

• Full documentation: https://docs.flowget.io
• Security policy: SECURITY.md
• License: FSL-1.1-ALv2 (see LICENSE)