About 5 min read

Autopilot Mode

Autopilot lets your avatars work autonomously on a schedule you define. Instead of waiting for your prompt, an avatar with Autopilot enabled executes tasks independently, reports results, and improves itself over time.

What Is Autopilot

In standard mode, an avatar responds only when you send a message. Autopilot reverses this dynamic: you define goals and a schedule, and the avatar takes the initiative.

| Aspect | Standard Mode | Autopilot Mode | |---|---|---| | Who initiates | You send a message | The avatar acts on schedule | | Frequency | On demand | Hourly, daily, weekly, or custom | | Results | Appear in conversation | Pushed to conversation + Telegram | | Improvement | You correct manually | Self-reflection every 10 tasks | | Label | (none) | "Avatar -- Proactive" badge |

Prerequisite: Autopilot is available on the Starter plan (limited) and Pro plan and above (full).

Setting Up Autopilot

Step 1: Open Autopilot Configuration

  1. Open the conversation with the avatar you want to automate.
  2. Tap the avatar name at the top to enter avatar settings.
  3. Find the Autopilot section.
  4. Tap Configure Autopilot.

Step 2: Define Goals

Goals are the core instructions that drive your avatar's autonomous work. Be clear and specific about what you want done.

Good goal examples:

Every morning, analyze the following 5 stocks: Apple, Google, Tesla, Nvidia,
and Microsoft. For each:
1. Previous day's price change and volume trends
2. Key technical signals (support/resistance levels)
3. Related news or announcements
4. Summary recommendation (hold/watch/buy opportunity)
Output as a concise table plus a brief written summary.

Every Monday, compile a weekly industry digest covering:
- Major AI releases and funding rounds
- Noteworthy open-source projects
- Regulatory and policy changes
Format as a weekly report with a one-line commentary for each item.

Monitor our main competitor's product page and track changes in:
- Pricing adjustments
- New feature releases
- Marketing copy changes
When changes are detected, generate a comparative analysis report.

Goal writing guidelines:

| Element | Description | Example | |---|---|---| | What to do | Define the task clearly | Analyze stock trends | | What to analyze | Specific dimensions to cover | Price change, volume, technical signals | | What to output | Expected output format | Table + written summary | | What to watch for | Constraints or special rules | Analysis only -- do not make buy/sell recommendations |

Tip: The more specific your goals, the better the avatar performs. Vague goals produce vague results.

Step 3: Configure the Schedule

Set how often and when the avatar should execute its tasks:

| Schedule Option | Description | Best For | |---|---|---| | Hourly | Execute every 1, 2, 4, 6, or 12 hours | Real-time monitoring, urgent tracking | | Daily | Execute at a specified time each day | Morning briefings, daily reports | | Weekly | Execute at a specified day and time each week | Weekly digests, periodic reviews | | Custom | Use a cron expression | Non-standard scheduling needs |

Configuration examples:

  • Every day at 8:00 AM: Ideal for daily briefing use cases
  • Every hour: Ideal for real-time monitoring
  • Every Monday at 9:00 AM: Ideal for weekly reports

Tip: For daily briefings, schedule them 30 minutes before your workday starts. This gives the avatar time to execute and have the results ready when you check in.

Step 4: Save and Enable

Once you have confirmed your goals and schedule, tap Enable Autopilot. The avatar will begin working at the next scheduled execution time.

Viewing Execution History

Every Autopilot run is saved in the conversation history.

In the Conversation

Messages generated by Autopilot are marked with a distinctive "Avatar -- Proactive" badge, clearly distinguishing them from your manual conversations.

You can:

  • Browse all automated execution results chronologically
  • Review the details of each run
  • Ask follow-up questions about any automated result
  • Compare outputs across multiple execution cycles

Execution Status

| Badge | Meaning | |---|---| | Avatar -- Proactive | Message generated by Autopilot | | Running | Task is currently executing | | Completed | Task finished successfully | | Failed | Task encountered an error (tap for details) |

Self-Evolution Mechanism

This is the most distinctive aspect of Autopilot: the avatar can optimize itself over time.

How It Works

After every 10 completed Autopilot tasks, the avatar triggers a reflection cycle:

  1. Reviews the last 10 task executions
  2. Analyzes what went well and what could be improved
  3. Summarizes lessons learned
  4. Applies improvements to subsequent tasks

Reflection Example

[Self-Reflection -- Iteration 10]

Review of last 10 daily stock analysis tasks:

What went well:
- Data accuracy remained consistently high
- Output format was uniform and easy to read

Areas for improvement:
- Insufficient attention to abnormal volume changes -- missed volume
  spike signals 3 times
- Technical analysis was too conservative -- should add MACD and RSI
  indicator interpretation

Improvement plan:
- Add anomalous volume detection in future analyses
- Include MACD and RSI indicators in the technical analysis section

What Self-Evolution Improves

| Aspect | Before Reflection | After Reflection | |---|---|---| | Output format | May be inconsistent across runs | Converges on the most useful format | | Information sources | May miss relevant sources | Learns which sources yield best results | | Relevance filtering | May include noise | Gets better at distinguishing signal from noise | | Tone and style | May drift over time | Stays aligned with the original intent |

You do not need to trigger reflections manually. The system runs them automatically after the 10th, 20th, 30th (and so on) completed tasks. You can review each reflection in the conversation log.

Telegram Push Notifications

When Autopilot completes a task, the results can be pushed directly to your Telegram account.

Requirements

  • Your Telegram account must be connected to Karma One via the Karma Bot (@karmabox7bot).
  • See the Quick Start Guide for binding instructions.

What Gets Pushed

When an Autopilot task completes:

  1. The full result is saved to the avatar's conversation in-app.
  2. A summary (or the full result if short enough) is sent to your Telegram chat with the Karma bot.
  3. You can read the notification on Telegram and optionally open the app for the full version.

Notification Settings

You can adjust notification behavior in the Autopilot configuration:

| Option | Description | |---|---| | Always notify | Push a notification after every task completion | | Notify on failure only | Only push when a task fails | | Off | No push notifications -- results saved in conversation only |

Morning Briefing

Morning Briefing is a specialized Autopilot mode designed for daily summaries.

Setting Up a Morning Briefing

  1. Open an avatar's conversation.
  2. Navigate to Morning Briefing settings (available in the conversation menu).
  3. Configure:
    • Time: When the briefing should be delivered (e.g., 7:30 AM)
    • Content scope: What topics to cover

What the Morning Briefing Includes

The briefing is generated by the avatar using its persona, knowledge base, and available tools (web search, calendar, etc.). A typical morning briefing might include:

  • Today's calendar events and meeting preparation notes
  • Overnight news relevant to your interests or industry
  • Task reminders and follow-up items
  • Market updates (if the avatar is configured for financial monitoring)

Morning Briefing vs. General Autopilot

| Feature | Morning Briefing | General Autopilot | |---|---|---| | Frequency | Once daily | Any schedule | | System prompt | Optimized for daily summary format | Uses your custom goal | | Telegram push | Automatic | Automatic | | Self-evolution | No | Yes (every 10 tasks) |

Proactive Engine

Beyond scheduled Autopilot, avatars can also operate in Proactive Mode, where they decide on their own when to reach out to you.

How Proactive Mode Works

  1. The system polls periodically (approximately every hour).
  2. A lightweight AI model evaluates whether there is something worth telling you, based on:
    • Recent events or news relevant to your interests
    • Changes in data the avatar monitors
    • Time-sensitive information
  3. If the AI decides to reach out, it generates a message and delivers it to your conversation and Telegram.
  4. A 4-hour cooldown prevents excessive notifications.

Enabling Proactive Mode

  1. Open the avatar's conversation.
  2. Navigate to Proactive Mode settings.
  3. Toggle it ON.
  4. Optionally configure what types of events should trigger a proactive message.

Tip: Proactive Mode works best for monitoring avatars (market watcher, news tracker, competitor analyst). For avatars that perform routine tasks, scheduled Autopilot is usually more appropriate.

Best Practices

Start Simple

Do not jump into complex tasks right away:

  1. Start with a simple daily task to test the setup
  2. Observe the results for a few days
  3. Adjust the goal description based on actual output quality
  4. Gradually increase task complexity

Make Goals Measurable

Good goal:

Every day, track our app's rating on the App Store and list the 5 most
recent user reviews.

Vague goal:

Keep an eye on how our product is doing.

Combine with Knowledge Bases

Autopilot combined with a knowledge base produces more expert automated tasks. For example, equip a customer support avatar with a knowledge base (product manual + FAQ), then enable Autopilot to summarize yesterday's customer inquiries daily. The result is an automated daily customer service report grounded in your actual product documentation.

Monitor Energy Consumption

Each Autopilot execution consumes energy. Be mindful of your balance when configuring high-frequency tasks.

| Frequency | Approximate Monthly Energy (Claude Sonnet 4) | |---|---| | Once daily | ~60-150 | | Three times daily | ~180-450 | | Once hourly | ~1,440-3,600 | | Once weekly | ~8-35 |

Review the First Few Results

After enabling Autopilot, actively review the initial 3-5 outputs and provide corrections if needed. This front-loaded investment pays off in long-term quality.

Managing Autopilot

Pause and Resume

On the Autopilot settings page, toggle the enable switch OFF to pause or ON to resume. Your configuration (goals, schedule) is preserved.

Modify Goals

You can update goals and schedules at any time. Changes take effect from the next scheduled execution.

Disable Autopilot

When you no longer need it, tap Disable Autopilot. The avatar reverts to normal mode, and all previously generated Autopilot conversation history is preserved.

Frequently Asked Questions

Will Autopilot run while I am asleep?

Yes. Autopilot executes according to its schedule regardless of whether you are online. If you set it for 2:00 AM, it will run at 2:00 AM.

What happens if an execution fails?

The system logs the failure reason, which you can review in the execution history. Common causes include insufficient energy and network issues. Failed tasks are not automatically retried; execution resumes at the next scheduled time.

Can multiple avatars have Autopilot enabled simultaneously?

Yes. Each avatar's Autopilot operates independently. Keep an eye on your total energy consumption.

Can I still chat with an Autopilot-enabled avatar normally?

Absolutely. Autopilot runs as background work and does not interfere with your ability to chat with the avatar at any time.

Plan Availability

| Feature | Free | Starter | Pro | Team | Enterprise | |---|---|---|---|---|---| | Autopilot | No | 3 active tasks | Unlimited | Unlimited | Unlimited | | Morning Briefing | No | 1 avatar | Unlimited | Unlimited | Unlimited | | Proactive Mode | No | No | Yes | Yes | Yes | | Self-evolution | N/A | Yes | Yes | Yes | Yes | | Telegram push | N/A | Yes | Yes | Yes | Yes |