Cron & Scheduling

Cron jobs are scheduled tasks that run your agents automatically. They are the primary way to make agents work without human intervention — monitoring systems, generating reports, delivering briefings, and triggering downstream workflows on a set cadence.

---

How cron jobs work

When a cron job fires:

1. AgenFleet opens a new isolated session for this job (separate from any chat sessions)
2. The agent’s SOUL file and memory context are loaded
3. Your prompt is delivered to the agent as the task
4. The agent executes — calling tools, reasoning, producing output
5. The output is delivered to your configured delivery channel
6. The isolated session is closed

The isolated session means cron job history never pollutes your interactive chat sessions with the agent. Each job run is clean and self-contained.

---

Creating a cron job

Navigate to your agent’s detail view → Settings tab → Cron Jobs → New Job.

Required fields

Name — A label for this job (e.g., “Daily Market Briefing”, “Weekly Compliance Check”).

Prompt — The task instruction delivered to the agent. See Writing effective prompts below.

Schedule — A cron expression defining when the job runs.

Delivery — Where to send the agent’s output.

Optional fields

Timezone — Defaults to UTC. Set this if your schedule is time-sensitive (e.g., “every morning at 8 AM your time”).

Enabled — Toggle to pause a job without deleting it.

---

Cron expression syntax

AgenFleet uses standard 5-field cron syntax:

┌───────────── minute (0–59)
│ ┌───────────── hour (0–23)
│ │ ┌───────────── day of month (1–31)
│ │ │ ┌───────────── month (1–12)
│ │ │ │ ┌───────────── day of week (0–6, Sunday=0)
│ │ │ │ │
* * * * *

Common schedules:

Description	Expression
Every day at 8 AM	0 8 * * *
Every Monday at 9 AM	0 9 * * 1
Every weekday at noon	0 12 * * 1-5
Every 6 hours	0 */6 * * *
Every hour	0 * * * *
First day of the month	0 9 1 * *
Every Sunday at 7 AM	0 7 * * 0

Tip

Use a cron expression validator (e.g., crontab.guru) to verify your schedule before saving. A misconfigured expression won’t error — it just won’t run when you expect.

---

Writing effective prompts

The prompt is what gets delivered to the agent as its task. The quality of the agent’s output is directly proportional to the quality of the prompt.

Good cron prompts are:

• Task-scoped — clearly define exactly what to produce
• Output-scoped — specify format, length, and structure
• Self-contained — the agent should be able to execute without clarification

Example — daily briefing:

Search for the top 3 AI industry news stories from the last 24 hours.

For each story, write:
- A one-sentence headline
- A 2-3 sentence summary
- Why it matters for enterprise AI adoption

Keep the total under 400 words. Format clearly for a Telegram message with bold headers.

Example — weekly health report:

Review the system health metrics from the past 7 days.
Identify any anomalies, trends, or items requiring attention.

Produce a structured report with:
1. Executive summary (3 sentences max)
2. Key metrics with week-over-week comparison
3. Issues requiring action (if any)
4. Recommended next steps

Use markdown formatting. Keep it under 500 words.

Avoid:

• Open-ended prompts like “Check on things and report back”
• Prompts that require clarification (“Should I include X?”)
• Prompts longer than ~300 words — keep the instruction tight

---

Delivery options

Every cron job output is delivered to a channel you configure. Multiple delivery targets are supported per job.

Telegram

Sends the agent’s output as a Telegram message. Supports direct messages and group channels.

• Requires a connected Telegram account under Settings → Notifications
• Long outputs are split into multiple messages automatically (Telegram’s 4096-char limit)
• Supports Markdown formatting in messages

Slack

Posts to a Slack channel or direct message.

• Requires a connected Slack workspace
• Output is posted as a bot message
• Supports @mentions if specified in the prompt

Email

Sends output as an email to any address.

• No pre-connection required — just enter the recipient email
• Subject line is auto-generated from the job name
• Output is rendered as plain text email body

Webhook

Posts a JSON payload to any HTTP endpoint.

{
  "jobId": "daily-briefing",
  "agentName": "Nova",
  "runAt": "2026-03-17T08:00:00Z",
  "output": "...",
  "tokensUsed": 1842,
  "status": "success"
}

Use webhooks to chain agents, trigger n8n workflows, or push output into your own systems.

Note

Delivery is controlled entirely by the job-level delivery configuration. The agent itself doesn’t control where output goes — this is intentional, so delivery can be changed without modifying the agent’s SOUL file or behavior.

---

Monitoring cron job runs

The agent’s Activity tab shows a log of every cron job execution:

• Job name and schedule
• Run timestamp
• Tokens consumed
• Delivery status (Delivered / Failed)
• Output preview (click to expand)

Failed deliveries are retried once automatically. If the second attempt fails, you receive an alert and the job is marked as failed for that run. The next scheduled run proceeds normally.

---

Managing jobs

Pause a job — toggle the Enabled switch. The job won’t run until re-enabled. Useful when you’re traveling or temporarily don’t need a briefing.

Edit a job — change prompt, schedule, or delivery at any time. Changes take effect on the next scheduled run.

Run now — trigger the job immediately without waiting for the schedule. Useful for testing new prompts or one-off runs.

Delete a job — removes the job permanently. Past run history is preserved in the activity log.