OpenClaw agents are coding agents that run 24/7, accessed via apps like Slack or Telegram, and are ideal for proactive work. Unlike on-demand tools like Claude Code, these agents continuously operate, freeing you from repetitive tasks and saving time and focus. This article covers setting up a fleet of OpenClaw agents and techniques for effective orchestration.

The initial setup is simple: prompt a Claude Code session to configure an agent on your environment, connect it to your Codex account, and integrate with Slack. Codex is recommended over Claude Code for this purpose. The real value lies in scaling and managing many agents efficiently.

Key strategies include building a shared knowledge base with a root markdown index, plus a bot-specific personal knowledge base updated via daily cron jobs. Cron jobs enable proactive behavior: examples include nightly "dreaming" sessions for agents, a daily QA bot that sends bug reports, and a finance bot that checks invoices. Be liberal with cron jobs—they are cheap and powerful.

Teaching agents tasks is essential. Each agent must be manually instructed on its specific duties, such as how to test an application or report results. This upfront investment ensures agents perform reliably and improve over time, creating a highly effective fleet.