OpenClaw, a fast-growing open-source AI agent platform, can generate surprisingly high API costs for everyday users. Key cost drivers include large system context files loaded on every call, growing conversation history, frequent heartbeat checks hitting expensive models, and lack of model routing. Practical strategies to reduce spending include consolidating multi-agent setups into a single agent with skills, routing simple tasks to cheaper or local models, enabling prompt caching for static context, trimming SOUL.md and conversation history, running local models (e.g., Qwen 3 32B via Ollama) for lightweight tasks, and using observability tools to track per-prompt costs. Users who applied these techniques reported cost reductions of 70–90%.