1. Enable Prompt Caching Prompt caching is like telling the AI provider: “I’m going to reuse this same instruction text a lot—please remember it so I don’t pay for it again every time.” Caching cuts costs by 50-90% on repeated context. Savings: ~$50-80/month Setup: Just ask. "Please set up prompt caching". 2. Truncate Session Context With daily use of Kimi 2.5, I noticed I made it through $70 in two weeks. I had to investigate. I found that I wasn't typing "/new" enough, and my sessions were long. A “213KB session” meant my conversation/history is huge.​The problem is: many systems resend the whole conversation history on every new message, so you keep paying for the same old text over and over.​ The fix in the notes is to automatically archive sessions once they get too big (example: over 50KB) and start fresh after a certain number of messages. Savings: ~$30-50/month Setup: Just ask. "Auto-archive sessions >50KB, force new sessions after N messages" 3. Model Routing by Task Use cheaper models for simple tasks: - Qwen 3.5: $0.40 /M tokens (coding, analysis) - Kimi K2.5: $0.60/M tokens (complex reasoning only) - Savings: ~$20-40/month *Disclaimer: This approach is in contention. Much of the economical reporting around real model usage varies based on what the model is actually doing. Some tasks require a certain context load, and it's hard to siphon that accurately without guidelines. It doesn't know what it doesn't know, and neither do you. Always budget with a cushion and monitor closely, this is a vehicle, it must be watched closely. I've shown you how to get this thing on the road, but under the hood, there's so much more to this lobster beast than how it feels on the road. There's the engine, the pipes, the radio, the steering wheel, and much more we will be working through. Welcome to The Open Campus for AI: An Academy for OpenClaw! --- Happy Chinese New Year! I hope that with this incoming eclipse season and year of the horse, you get to see the fruits of your labor in fulfilment and abundance 🎈🐎