Start

A proactive assistant has two halves. The generating half decides what is worth saying. The suppressing half decides whether to say it now, later, or never. This package is the second half.

npm install proactive-gate

Zero dependencies, TypeScript, Node 22 or newer. For Python see Python.

import { createGate, defaultChecks, MemoryStore } from "proactive-gate";

const gate = createGate({
  store: new MemoryStore(),
  checks: defaultChecks({ dailyLimit: 3, quietHoursFloor: "high" }),
});

const user = {
  id: "ayse",
  consent: true,
  timezone: "Europe/Istanbul",
  quietHours: { start: "22:00", end: "08:00" },
};
const candidate = { id: "a1", type: "reminder", priority: "normal" };
const now = new Date("2026-09-04T00:00:00Z"); // 03:00 in Istanbul

const decision = await gate.evaluate({ user, candidate, now });
console.log(decision.allowed, decision.rejectedBy);
console.log(decision.reason);
console.log(decision.nextAllowedAt?.toISOString());

It prints:

false quietHours
quiet hours 22:00 to 08:00 (thu 2026-09-03) Europe/Istanbul; priority normal is below the floor (high)
2026-09-04T05:00:00.000Z

evaluate runs the checks in order and stops at the first reject. In production, leave now out and call gate.commit(decision, input) right before you send: it is the atomic increment of the budget, it returns false when a concurrent delivery took the last unit, and it is idempotent on decision.id. Use RedisStore, SqliteStore or PostgresStore instead of MemoryStore when more than one process sends.

The input

const user = {
  id: "ayse",
  consent: true,
  timezone: "Europe/Istanbul",
  quietHours: { start: "22:00", end: "08:00" },
  createdAt: "2026-01-01T00:00:00Z",
};
const candidate = { id: "a1", type: "reminder", priority: "normal", surfaces: ["push", "feed"] };

A user needs id and consent; everything else is optional and each check says what it reads. A candidate needs id and type; priority defaults to normal. The gate never reads payload.

Replay a day before you ship a policy

This chain needs nothing but the package: simulate writes a generated week of candidates, init writes a policy, and replay runs the one on the other.

npx proactive-gate simulate --dump-events events.jsonl
npx proactive-gate init
npx proactive-gate replay events.jsonl --policy proactive-gate.policy.json --commit

The CLI takes a JSONL file of { user, candidate, now } lines and prints the allow rate and the top rejection reasons. --json prints one decision per line for a notebook. Point it at your own candidates instead of the generated week to see what a policy would have done to them.