Run Scheduled Tasks With Zero Tokens Using a System Timer

Why do it this way

Many people have an Agent do recurring work every day: pull data, compile reports, back up files, check websites. If the Agent runs every time, each run costs Tokens, which adds up to a lot over a month, and the Agent's execution may not be the same each time.

A better approach:

• Have the Agent do the work once: write the task as a Python script. This costs some Tokens, one time.
• Then hand the script to the system's built-in scheduler. It doesn't go through the AI, so each run costs 0 Tokens.
• When the script finishes, it messages you the result, so you know whether it succeeded or failed.

The result: stable execution, predictable output, and almost no cost.

Step 1: Have the Agent write the task as a script

Say this to your Agent (Claude Code, Codex, etc.):

The script it produces will look roughly like this:

import os, json, time, urllib.request, urllib.parse

def notify(text):
    token = os.environ["TG_BOT_TOKEN"]
    chat = os.environ["TG_CHAT_ID"]
    data = urllib.parse.urlencode({"chat_id": chat, "text": text}).encode()
    urllib.request.urlopen(f"https://api.telegram.org/bot{token}/sendMessage", data, timeout=10)

def run():
    start = time.time()
    with urllib.request.urlopen("https://example.com", timeout=15) as r:
        return r.status, round(time.time() - start, 2)

try:
    status, seconds = run()
    notify(f"✅ Website check complete: status {status}, took {seconds} seconds")
except Exception as e:
    notify(f"❌ Website check failed: {e}")

Run it manually once first to confirm you receive the message:

python3 /full/path/task.py

Step 2: Hand it to the system to run on a schedule

macOS / Linux: use cron

crontab -e

Add a line (runs every day at 9 AM):

0 9 * * * TG_BOT_TOKEN=your_token TG_CHAT_ID=your_id /usr/bin/python3 /full/path/task.py >> /full/path/task.log 2>&1

The first five fields are, in order: minute, hour, day, month, weekday. 0 9 * * * means every day at 9:00.

Windows: use Task Scheduler

schtasks /create /tn "DailyTask" /tr "python C:\full\path\task.py" /sc daily /st 09:00

Set the environment variables in "System Properties → Environment Variables".

Step 3: Confirm it is actually running

• The next day, check whether the message arrived on your phone.
• If not, check the errors in task.log.
• Use the full path in cron commands: cron's environment is very bare, so python3 and your file paths may not be found. Use which python3 to find Python's full path.
• On macOS, if the script reads or writes certain folders (such as Desktop or Documents), you may need to grant cron "Full Disk Access".

When you still need AI

• When the task requires understanding content: for example, summarizing an article or deciding how to reply to an email. Only that part calls the model; the other steps still use the script.
• When requirements change: have the Agent update the script once, then keep running it at 0 Tokens.

If a task can be written as fixed steps, don't let AI improvise it every time.

Further reading

• For why Agent workflows should be scripted, see the experience piece OpenClaw's fatal problem.
• For more complete message push setups, see the Workflow handbook.