dabke TypeScript DSL for OR-Tools CP-SAT workforce scheduling

[christianklotz]

I've open-sourced https://github.com/christianklotz/dabke, a TypeScript library that provides a declarative DSL for workforce scheduling on top of CP-SAT.

The core idea: Instead of writing individual low-level OR-Tools constraint implementations for each scheduling problem, dabke gives you a higher-level abstraction. You describe employees, shift patterns, coverage requirements, and composable rules. The ModelBuilder compiles all of that down to a CP-SAT model and sends it to the solver.

import { ModelBuilder, HttpSolverClient } from "dabke";

const builder = new ModelBuilder({
  employees: [
    { id: "alice", roleIds: ["server"] },
    { id: "bob", roleIds: ["server"] },
  ],
  shiftPatterns: [
    { id: "morning", startTime: { hours: 8 }, endTime: { hours: 16 } },
    { id: "evening", startTime: { hours: 16 }, endTime: { hours: 23 } },
  ],
  coverage: [
    { day: "2026-02-09", startTime: { hours: 8 }, endTime: { hours: 16 },
      roleIds: ["server"], targetCount: 1, priority: "MANDATORY" },
    { day: "2026-02-09", startTime: { hours: 16 }, endTime: { hours: 23 },
      roleIds: ["server"], targetCount: 1, priority: "MANDATORY" },
  ],
  schedulingPeriod: { specificDates: ["2026-02-09"] },
  ruleConfigs: [
    { name: "max-hours-day", config: { hours: 8, priority: "MANDATORY" } },
    { name: "min-rest-between-shifts", config: { hours: 10, priority: "MANDATORY" } },
  ],
});

const { request } = builder.compile();
const solver = new HttpSolverClient(fetch, "http://localhost:8080");
const response = await solver.solve(request);

How it works: Each rule implements a compile(builder) method that emits constraints into the model. Rules are composable and scopeable, so you can target a constraint at specific employees, roles, skills, days of the week, or date ranges, or a combination of which. Adding a new scheduling constraint means writing a new rule, not reworking the model.

The rule system is extensible — you can write custom rules that hook into the same builder API. It also supports semantic time (named periods like "lunch" that vary by day of week) and pre/post-solve validation.

It comes with a Docker setup for getting started quickly.

[jmarca]

Nice! I’ve done some scheduling work with ORTools too, so I’ll give a look when I get some time. From the readme, it looks promising.