11 — Heartbeat / Automation Workflows

What Was Researched

Scheduled, recurring, and event-driven automation patterns in AI agents — cron jobs, background tasks, webhooks, and unattended operation.

Which Sources Were Used

Source Type URL Relevance
Hermes Agent (hermes-agent/cron/, tools/cronjob_tools.py) Local codebase https://github.com/NousResearch/hermes-agent CRITICAL
OpenClaw automation docs Local codebase https://github.com/openclaw/openclaw HIGH

Key Findings

Hermes Cron System

  • cron/ directory — jobs.py + scheduler.py for scheduled task management
  • tools/cronjob_tools.py (47KB) — Agent-accessible cron tool
  • Natural language scheduling — "Every Monday at 9am, send me a weekly summary"
  • Platform delivery — Cron results delivered to any connected platform (Telegram, Slack, etc.)
  • Unattended operation — Runs while user is away
  • Gateway integration — Cron jobs run through the messaging gateway for platform delivery

OpenClaw Automation

  • Cron jobs — Scheduled tasks documented as a core feature
  • Webhooks — automation/webhook endpoint for event-driven triggers
  • Gmail Pub/Sub — Email-driven automation via Google Pub/Sub
  • Gateway-managed — All automation runs through the gateway control plane

Hermes Batch Processing

  • batch_runner.py (59KB) — Parallel batch processing for multiple tasks
  • Trajectory generation — Automated generation of agent interaction trajectories for training data
  • trajectory_compressor.py (70KB) — Compresses trajectories for model training

Automation Patterns

Pattern Hermes OpenClaw Description
Cron scheduling ✅ ✅ Time-based recurring tasks
Webhook triggers ❌ ✅ Event-driven via HTTP
Gmail Pub/Sub ❌ ✅ Email-triggered
Batch processing ✅ ❌ Parallel multi-task
Trajectory gen ✅ ❌ Training data pipeline
Platform delivery ✅ ✅ Results to messaging

What Is Confirmed

  1. Cron scheduling is the baseline — both Hermes and OpenClaw implement it
  2. Platform delivery is essential — results must reach the user wherever they are
  3. Natural language scheduling improves UX dramatically
  4. Webhook/event-driven automation enables integration with external systems
  5. Batch processing is valuable for research and data-processing workflows

What Is Uncertain

  • Optimal scheduling persistence (database vs. filesystem)
  • How to handle cron job failures (retry, alert, or skip)
  • Whether to support event-driven triggers beyond webhooks

How This Applies to Building a Modern Model-Agnostic Agent Harness

  1. Implement cron scheduling — time-based recurring tasks
  2. Support platform delivery — route results to any connected channel
  3. Natural language scheduling — let the agent parse "every Monday at 9am"
  4. Consider webhook support — enables external system integration
  5. Implement batch processing — useful for parallel task execution