From $36K to $10K: The 5 OpenClaw Settings That Halve Your Bill

I made a video on cutting your OpenClaw bill, and here is the full playbook so you do not have to pause and scrub through it. A team of 10 developers was on track to spend about $36,000 a year. Five settings later that number drops to around $5,000 to $10,000. The same five fixes cut my own bill by half. None are complicated, and they work in a specific order.

Start with your subscription, then the heartbeat

The first fix is the one I got wrong myself. I set up OpenClaw against my raw Claude API key and burned about $38 in a few minutes. If you already pay $20 a month or more for ChatGPT or Codex, you can connect OpenClaw to that subscription instead of paying per token. OpenClaw’s docs say this is officially supported for Codex. Claude Code banned the same trick, so Codex is the safe route for now.

The second fix is the heartbeat, a background process that checks in with the model every 30 minutes to ask if anything changed. The catch: each check-in sends your entire project setup, so people estimate it adds $50 to $150 a month on an expensive model for doing nothing. In your settings file you can turn it off, point it at a cheap model, limit it to work hours, or switch on isolated sessions so each check-in starts fresh. The docs say that last change drops a check-in from about 100,000 tokens to 2,000 to 5,000.

The other three

• Route by task. Most people run one expensive model for everything. Send status checks and simple config changes to a cheap model, everyday work to a mid model, and save the frontier model for real reasoning like architecture or security review.
• Stop re-billing context. Every message resends your project setup, tool list, and full history. The docs show 13,000 tokens of background going out before your question even lands. Type /new when you switch tasks instead of running one thread for hours.
• Schedule your server. If OpenClaw runs on a cloud VM, it bills 24/7 even though you use it maybe 8 to 10 hours on weekdays. One command starts it in the morning, one stops it at night. Skip this if you run it on your own machine or genuinely need it up for cron jobs.

Why the order matters

Doing them in sequence means you capture the biggest savings first, so even if you stop after subscription auth and the heartbeat fix you have already moved the number a lot. The team example scales the exact same moves up to five figures.

Watch the full playbook, fix by fix, in the video above. For more ways to run OpenClaw cheaply, see the related guides on openclawdc.com. I specialize in cutting AI and cloud bills like this for teams. Book a call at cloudyeti.io/meet.