How to Build an AI Agent That Runs on a Schedule (Cron, GitHub Actions, Cloudflare)

Intermediate ~6 min read Last tested October 2026

Updated 2026-10-15

Many useful agents aren’t chatbots at all: a morning inbox digest, an hourly price checker, a weekly report. This guide shows three ways to run an agent on a schedule, from simplest to most serverless.

Prerequisites

An agent script that runs end to end from the command line (e.g. from Build your first AI agent in Python)
An LLM API key

Make the agent schedule-friendly first

A scheduled agent runs with no one watching, so:

Hard-cap steps and tokens (max_steps, max_tokens).
Make it idempotent: record what was already processed (a file, a DB row) so a re-run doesn’t double-send.
Send the result somewhere (email, Slack, a file) and log failures loudly.

# daily_agent.py
import datetime, json, pathlib
from agent import run_agent          # your agent loop

state = pathlib.Path("state.json")
seen = json.loads(state.read_text()) if state.exists() else {}
today = datetime.date.today().isoformat()
if seen.get("last_run") == today:
    raise SystemExit("already ran today")

summary = run_agent("Summarize today's top 5 Hacker News stories for a developer.", max_steps=6)
print(summary)                        # or post to Slack / email
state.write_text(json.dumps({"last_run": today}))

Option 1: cron on a server or Mac/Linux machine

crontab -e

# every day at 07:30 server time
30 7 * * * cd /home/me/agent && /home/me/agent/.venv/bin/python daily_agent.py >> agent.log 2>&1

Cron does not load your shell profile. Use absolute paths and put API keys in an env file the script loads (e.g. with python-dotenv).

Option 2: GitHub Actions (free, no server)

# .github/workflows/agent.yml
name: daily-agent
on:
  schedule:
    - cron: "30 14 * * *"     # 14:30 UTC daily
  workflow_dispatch: {}        # lets you run it by hand
jobs:
  run:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with: { python-version: "3.12" }
      - run: pip install -r requirements.txt
      - run: python daily_agent.py
        env:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}

Add the key under Settings → Secrets and variables → Actions. Schedules are in UTC, can start several minutes late at busy times, and are disabled automatically in public repos with no activity for 60 days.

Option 3: Cloudflare Workers Cron Triggers (serverless, JavaScript)

# wrangler.toml
name = "scheduled-agent"
main = "src/index.js"
compatibility_date = "2026-10-01"

[triggers]
crons = ["30 14 * * *"]

// src/index.js
export default {
  async scheduled(controller, env, ctx) {
    const r = await fetch("https://api.openai.com/v1/chat/completions", {
      method: "POST",
      headers: { "Authorization": `Bearer ${env.OPENAI_API_KEY}`, "Content-Type": "application/json" },
      body: JSON.stringify({
        model: "gpt-4.1-mini",
        messages: [{ role: "user", content: "Write a one-line motivational note for a developer." }],
      }),
    });
    const data = await r.json();
    console.log(data.choices[0].message.content);   // or POST to a Slack webhook
  },
};

npx wrangler secret put OPENAI_API_KEY
npx wrangler deploy
npx wrangler dev --test-scheduled   # then: curl "http://localhost:8787/__scheduled?cron=30+14+*+*+*"

Workers have CPU-time limits; network waiting on the LLM doesn’t count toward CPU time, but long multi-step agents may suit a Workflow or a queue better.

Common errors

Cron job does nothing: check agent.log; usually python: command not found or a missing env var. Use absolute paths.

GitHub Action never fires: the workflow file must be on the default branch, and scheduled runs pause after 60 days of repo inactivity.

Runs at the wrong hour: GitHub Actions and Cloudflare cron use UTC.

Agent sends the same report twice: add the idempotency check shown above.