An OpenClaw skill that reduces API costs by 50-80% through intelligent model routing, context compression, and smart optimization.
- 🧠 Intelligent Model Routing — Automatically classifies question complexity and switches to the most cost-effective model
- 📦 Context Compression — Summarizes long conversations to reduce input tokens
- ✂️ Reply Efficiency — Matches response length to question complexity
- ⚡ Tool Call Optimization — Combines commands, eliminates redundant operations
User: "你好" (simple greeting)
→ Agent classifies as Level 1 → Switches to Haiku ($0.80/1M)
User: "帮我设计一个系统架构" (complex task)
→ Agent classifies as Level 3 → Switches to Opus ($15/1M)
User: "好的谢谢" (simple reply)
→ Agent classifies as Level 1 → Switches back to Haiku
70% of daily messages are simple/medium → saves 60-80% compared to always using the strongest model.
| Optimization | Cost Reduction |
|---|---|
| Intelligent model routing | 60-80% on simple/medium tasks |
| Context compression (8+ turns) | 40-60% input reduction |
| Filler elimination | ~20-50 tokens/reply |
| Tool call reduction | ~500-2000 tokens each |
| Combined commands | ~40% tool overhead reduction |
Total: 50-80% monthly API cost reduction
openclaw skills install token-saverOr manually copy SKILL.md to your OpenClaw skills directory:
~/.openclaw/workspace/skills/token-saver/SKILL.md
Once installed, the skill is always active. It teaches the agent to:
- Compress context after 8+ conversation turns instead of carrying full history
- Write to files instead of keeping important info only in conversation memory
- Respond concisely — no filler phrases, length matched to complexity
- Optimize tool usage — combine commands, skip redundant reads, batch operations
Issues and PRs welcome!
MIT