Scheduled agent tasks
Set your AI agent to check in automatically on a schedule — daily reminders, weekly reviews, or conditional check-ins only when action is needed.
Your agent can do more than reply to you in chat — you can set it to check in on a schedule and message you proactively. Think of it as a personal coach that taps you on the shoulder at a set time, or a researcher that brings you the news digest every morning, or a habit tracker that reminds you only when you've forgotten to log something.
Creating a scheduled task
You have two ways to set one up.
In chat — just ask. Tell your agent what you'd like: "Send me a motivational message every morning at 8" or "check if I've logged calories after lunch and remind me only if I haven't". Your agent will confirm the task is set, and it'll start running on schedule.
In the app — open Automations in the sidebar. Every rule you have is listed there; the scheduled ones carry a Scheduled badge and show their last run. New automation creates one, and Edit on a rule opens its JSON — where the agent, the instruction and the schedule all live.
Schedule options
When you create or edit a task, pick how often it runs:
- Daily at a time — every day at the same time (e.g. "8:00 AM").
- Weekly on specific days — pick the days of the week and a time (e.g. "Monday and Thursday at 6 PM").
- On an interval — every 30 minutes up to every 24 hours. The next run is counted from the previous one.
All times are in your timezone — Kaizendex uses the time zone on your device, so if you travel the schedule travels with you.
How the message arrives
When a scheduled task runs, the result appears as a message in that agent's chat. You'll also get a notification (the Inbox icon at the bottom of the sidebar's rail shows a badge) — if you've enabled push notifications on this device, you'll get notified even if the app is closed.
A task runs at its scheduled time, and if it produces a message, the message lands right away. The notification is written down the moment the run finishes and delivered by a background worker that retries on its own — so even if something hiccups on our side, "your agent replied" still reaches you a minute or two later instead of getting lost.
Stopping a run
While a task is running, its row on the Automations page shows a Running… badge. If you don't want to wait for it — it's taking too long, or you changed your mind — open the row's menu (or, on the phone, tap Stop on the row) and choose Stop run.
- A run that hasn't started yet is cancelled straight away.
- A run that is already working stops at its next safe point, usually within a few seconds. Anything it had already saved stays saved; nothing half-done is left behind.
A stopped run shows as cancelled in the task's history. It does not count as a failure, so stopping a task never switches it off.
If a server goes down in the middle of a run, Kaizendex notices within a few minutes, marks the run as interrupted and sends you a notification saying so — the task simply runs again at its next scheduled time.
Scheduled runs use the same AI model as your chats — the plan you picked in Settings → AI settings → Model — so a scheduled check-in is exactly as sharp as talking to the agent directly.
Silent when there's nothing to say
Some tasks are conditional — they check whether something needs doing and only message you if the answer is yes. For example, "Remind me to log calories only if I haven't already." If you have logged calories, the agent finds nothing to report and the task runs silently — no message, no notification, no trace.
This keeps your inbox clean: you hear from the agent only when there's something that actually needs your attention.
The weekly tune-up
Once an agent has a scheduled task of yours, it also gets a Weekly tune-up — a normal task (Sundays at 18:00, in your Tasks list) where the agent reviews itself: did it claim things your data doesn't support, is there data of yours it never looks at, does it actually know what a "good day" means to you, and did you correct it in chat without that correction being saved anywhere?
When it finds something, it messages you a concrete proposal — for example a draft definition of a productive day based on your real numbers — and asks for your approval before changing anything about how it works. It never rewrites its own instructions silently.
The tune-up is yours like any other task: reschedule it, edit its instructions, turn it off, or delete it — a deleted tune-up stays deleted.
Managing your tasks
Open AI → Tasks to see all your scheduled tasks for all your agents. Each task card shows:
- The agent running it (its avatar and name).
- The schedule in plain English ("Daily · 8:00 AM", "Mon, Thu · 6:30 PM").
- When it last ran and whether it succeeded.
- An on/off switch to pause and resume it without deleting it.
Tap the three-dot menu to Edit (change the schedule or instructions) or Delete (remove it).
You can have up to 10 active scheduled tasks at a time.
What if you delete an agent?
If you remove an agent that has scheduled tasks, those tasks are also removed. If you change your mind and undo the deletion (you'll see an Undo button), the agent and its tasks come back.
Tasks that fail repeatedly
Scheduled runs are monitored automatically. If a task fails to complete for 5 times in a row, Kaizendex turns it off automatically to save resources and sends you a system notification. You can re-enable it in AI → Tasks once the issue is fixed, or delete it and create a new one.
Research tasks with web access
Every agent can search the web, in chat and in scheduled tasks. For example, "Read Hacker News every morning and message me the top 5 interesting things" — the agent searches and reads the web during the run and sends you the summary as its message. The pages it used show as links under the reply.
When a long task reaches its limit
A long background task can stop when it reaches its token or estimated cost allowance. It keeps the work already saved and tells you which limit it reached. Restarting its worker does not reset the allowance, and help from another model counts toward the same total.
For example, a research task may save the sources it has already read and stop before preparing the full report. Its message explains that the task is incomplete.
Next
- Notifications — where notifications appear and how to enable push
- Chats and AI agents — how to talk to your agents