Scheduled Tasks

Natural language scheduling - agents that work on a schedule

Ask ChatGPT Ask Claude.ai Open in Cursor

Schedule recurring tasks using natural language. Say "check my email every hour" or "remind me Monday at 9am" and agents execute those tasks automatically, on schedule.

What Are Scheduled Tasks?

Scheduled tasks transform one-time requests into recurring agent actions:

User:  "Check my email every hour for messages from my boss"
         ↓
System creates scheduled task
         ↓
Every hour: Agent checks email, reports findings

The magic: No cron syntax, no configuration files. Just natural language.

How It Works

Natural Language → Schedule

User:  "Remind me every weekday at 2pm to review open issues"
         ↓
System understands:
  - Frequency: Every weekday
  - Time: 2pm (user's timezone)
  - Action: Review open issues
         ↓
Creates cron: 0 14 * * 1-5
         ↓
Executes automatically

The system parses your intent and creates the schedule automatically.

Examples

"Check Slack every 30 minutes"
→ Runs every 30 minutes

"Send me a summary every Monday morning"
→ Runs Mondays at 9am (default morning time)

"Every 2 weeks on Tuesday, analyze metrics"
→ Runs every 2 weeks on Tuesday

"Daily at 5pm except weekends"
→ Runs weekdays at 5pm

"First Monday of each month at 10am"
→ Runs first Monday at 10am

No cron syntax needed - the agent figures it out.

Set your timezone in the formation. Scheduled tasks use the formation's configured timezone. Without it, times default to UTC, which may not match user expectations.

Creating Scheduled Tasks

Through Conversation

Just ask:

User:  "Check my GitHub notifications every hour"
Agent: "I've created a scheduled task to check your GitHub notifications
        every hour. Would you like me to start it now?"
User:  "Yes"
Agent: "Started! I'll check notifications every hour and notify you of
        anything important."

That's it. The schedule is active.

Verify Creation

User:  "Show my scheduled tasks"
Agent: "You have 3 scheduled tasks:
        1. Check GitHub notifications (every hour)
        2. Daily standup summary (weekdays at 9am)
        3. Weekly metrics report (Mondays at 10am)"

Modify Schedule

User:  "Change the GitHub check to every 2 hours"
Agent: "Updated! Now checking GitHub notifications every 2 hours."

Natural language all the way.

What Gets Scheduled

Commands & Actions

"Run tests every night at midnight"
"Deploy to staging every weekday at 3am"
"Backup database daily at 2am"
"Clean up old logs every Sunday"

Agents execute these actions on schedule with full tool access.

Reports & Summaries

"Email me a daily summary of customer feedback"
"Send weekly sales report every Friday"
"Monthly expense summary on the 1st"
"Quarterly OKR review reminder"

Agents gather data, analyze, and deliver reports automatically.

Monitoring & Alerts

"Check API health every 5 minutes"
"Monitor server disk space hourly"
"Alert me if error rate exceeds 1%"
"Daily security scan at midnight"

Proactive monitoring that alerts you only when needed.

Reminders & Notifications

"Remind me to review PRs every morning"
"Notify team of standup 10 minutes before"
"Send birthday wishes to customers"
"Weekly team sync reminder on Monday"

Never forget important tasks.

Timezone Support

Schedules respect user timezones:

User in New York: "Check email at 9am"
→ Executes 9am ET

User in London: "Check email at 9am"
→ Executes 9am GMT

Each user's schedule runs in their timezone.

Changing Timezones

User:  "I'm traveling to Tokyo next week"
Agent: "Would you like me to adjust your scheduled tasks to Tokyo time?"
User:  "Yes"
Agent: "Updated all schedules to JST timezone."

User Isolation

Each user has their own schedules:

Alice: "Check my email every hour"
Bob: "Check my email every hour"

Alice's schedule: Checks Alice's email
Bob's schedule: Checks Bob's email

Complete isolation - no data leakage.

Complex Schedules

Exclusions

"Every day except holidays"
→ Skips holidays automatically

"Every weekday except company holidays"
→ Skips weekends and holidays

"Every hour except 2am-6am"
→ Runs hourly, skips overnight

Conditional Execution

"Check Slack every hour during business hours"
→ Only runs 9am-5pm weekdays

"Send report if there's new data"
→ Only executes when data changes

Dynamic Rescheduling

User:  "If the report fails, retry in 30 minutes"
Agent: Creates schedule with retry logic

Managing Scheduled Tasks

List All Tasks

User:  "What are my scheduled tasks?"
Agent: "You have 5 active schedules:
        1. GitHub notifications (every hour) - Next: 2:00pm
        2. Daily summary (9am weekdays) - Next: Tomorrow 9am
        3. Weekly report (Monday 10am) - Next: Monday 10am
        4. API health check (every 5min) - Next: 1:35pm
        5. Monthly backup (1st at midnight) - Next: Feb 1"

Pause/Resume

User:  "Pause the GitHub notifications while I'm on vacation"
Agent: "Paused GitHub notifications. Enjoy your vacation!"

[Later]
User:  "Resume GitHub notifications"
Agent: "Resumed. Next check at 3pm."

Delete Tasks

User:  "Cancel the weekly report"
Agent: "Deleted weekly report schedule."

Modify Tasks

User:  "Change API health check to every 10 minutes"
Agent: "Updated API health check frequency to every 10 minutes."

Programmatic Management (CLI & SDK)

Developers can manage scheduled tasks programmatically:

# List all scheduled tasks for a user
muxi schedules list --user alice@acme.com

# Pause a scheduled task
muxi schedules pause github-notifications --user alice@acme.com

# Resume a scheduled task
muxi schedules resume github-notifications --user alice@acme.com

# Delete a scheduled task
muxi schedules delete weekly-report --user alice@acme.com

# Update schedule frequency
muxi schedules update api-health --cron "*/10 * * * *" --user alice@acme.com

from muxi import Muxi

client = Muxi()

# List all scheduled tasks for a user
tasks = client.schedules.list(user_id="alice@acme.com")

# Pause a scheduled task
client.schedules.pause("github-notifications", user_id="alice@acme.com")

# Resume a scheduled task
client.schedules.resume("github-notifications", user_id="alice@acme.com")

# Delete a scheduled task
client.schedules.delete("weekly-report", user_id="alice@acme.com")

# Update schedule frequency
client.schedules.update(
    "api-health",
    user_id="alice@acme.com",
    cron="*/10 * * * *"
)

import { Muxi } from '@muxi/sdk';

const client = new Muxi();

// List all scheduled tasks for a user
const tasks = await client.schedules.list({ userId: 'alice@acme.com' });

// Pause a scheduled task
await client.schedules.pause('github-notifications', { userId: 'alice@acme.com' });

// Resume a scheduled task
await client.schedules.resume('github-notifications', { userId: 'alice@acme.com' });

// Delete a scheduled task
await client.schedules.delete('weekly-report', { userId: 'alice@acme.com' });

// Update schedule frequency
await client.schedules.update('api-health', {
  userId: 'alice@acme.com',
  cron: '*/10 * * * *'
});

Behind the Scenes

Cron Generation

Natural language converts to cron expressions:

"Every day at 9am"          → 0 9 * * *
"Every weekday at 2:30pm"   → 30 14 * * 1-5
"First Monday at 10am"      → 0 10 1-7 * 1
"Every 6 hours"             → 0 */6 * * *

You never see the cron - it just works.

Intelligent Rephrasing

The system optimizes scheduled prompts:

User's original: "Can you check my email for messages from John?"
Scheduled prompt: "Check email for messages from John and report findings"

More concise, more effective for recurring execution.

Executes As The User

At execution time, the scheduled task is sent to MUXI as the user who created it:

User (alice@acme.com): "Send me a daily summary at 9am"
         ↓
Every day at 9am:
  Request sent as alice@acme.com
  Uses Alice's credentials
  Uses Alice's memory/context
  Response delivered to Alice

This means:

User's credentials are used for tool access
User's memory and synopsis are available
Results are isolated to that user

Context Preservation

Scheduled tasks maintain user context:

User:  "Use my work email, not personal"
[Later, scheduled task runs]
Agent: [Uses work email as instructed]

Preferences persist across scheduled executions.

Configuration

Enable Scheduler

# formation.afs
scheduler:
  enabled: true
  check_interval_minutes: 1    # Check for due tasks every minute
  timezone: "America/New_York" # Default timezone

Database Required

Scheduled tasks require persistent storage:

memory:
  persistent:
    enabled: true
    connection_string: ${{ secrets.POSTGRES_URI }}
    # Or: "sqlite:///./scheduler.db"

Example Use Cases

Development Workflow

"Run tests every time I push to GitHub"
"Deploy to staging every weekday at 3am"
"Send me a summary of new issues every morning"
"Remind me to review PRs every afternoon"

Operations & Monitoring

"Check API health every 5 minutes"
"Monitor disk space every hour"
"Backup database daily at 2am"
"Weekly security scan on Sunday"

Business Intelligence

"Daily sales summary at 8am"
"Weekly customer feedback report on Friday"
"Monthly expense analysis on the 1st"
"Quarterly revenue forecast on the 15th"

Personal Productivity

"Remind me to stretch every 2 hours"
"Daily standup reminder 5 minutes before"
"Weekly review prompt every Friday"
"Monthly goals check-in"

Advanced Features

One-Time Scheduled Tasks

"Remind me tomorrow at 3pm to call the dentist"
→ Runs once at 3pm tomorrow, then deletes

"Send report next Monday at 10am"
→ One-time execution

Chained Tasks

"Every night: backup database, then clean logs, then send report"
→ Executes in sequence automatically

Conditional Schedules

"Check deployment status every 10 minutes until it succeeds"
→ Stops automatically when condition met

Monitoring & Audit

Execution History

User:  "Show me the last 5 executions of my daily summary"
Agent: "Last 5 executions:
        1. Today 9:00am - Success (2 new items)
        2. Yesterday 9:00am - Success (5 new items)
        3. Monday 9:00am - Success (3 new items)
        4. Friday 9:00am - Success (7 new items)
        5. Thursday 9:00am - Failed (API timeout)"

Failure Handling

When a scheduled task fails:

Agent: "Your daily summary failed to run at 9am due to API timeout.
        I'll retry in 30 minutes. Would you like me to adjust the
        retry schedule?"

Automatic retry with intelligent backoff.

Success Notifications

User:  "Notify me when scheduled tasks succeed"
Agent: "I'll send you a notification after each successful execution."

[Later]
Agent: "✓ Daily summary completed successfully. Found 3 new items."

Why This Matters

Without Scheduled Tasks	With Scheduled Tasks
Manual cron configuration	Natural language
Remember to check things	Agent checks for you
Reactive (respond to events)	Proactive (anticipate needs)
One-time commands	Recurring automation
Setup scripts and jobs	Just ask

The result: agents that work while you sleep, not chatbots waiting for input.

Quick Start

# Enable scheduler
scheduler:
  enabled: true

memory:
  persistent:
    connection_string: ${{ secrets.POSTGRES_URI }}

Then just ask:

User:  "Check my email every hour for urgent messages"
Agent: "Done! I'll check your email hourly and notify you of anything urgent."

Learn More

Triggers & Webhooks - Real-time webhook-based automation
Tools & MCP - How agents use tools in scheduled tasks
Memory System - How context persists across scheduled executions

We use cookies