Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Routes OpenClaw agent requests to Haiku by default, escalating to Sonnet or Opus only when needed. Offloads heartbeat checks to local providers like Ollama or free tiers like Groq, reduces session context from ~50KB to ~8KB, and caches repeated prompts at 10% of standard cost. All changes write to ~/.openclaw/ with automatic backups.
Every change runs in dry-run mode by default, creates timestamped backups before writing, and produces a verification report after applying, so there is no guesswork about what changed or whether it worked.
When to use it
- Cutting a $1,500/month AI bill down to under $50
- Running heartbeat checks locally with Ollama instead of paying per token
- Previewing config changes in dry-run mode before applying them to a live setup
- Rolling back to a previous config after an optimization breaks agent behavior
- Setting daily and monthly token budget caps to avoid unexpected charges
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: optimize my OpenClaw setup to reduce API costs
- 1Runs `python cli.py analyze` to inspect current model routing, heartbeat config, and context load
- 2Runs `python cli.py optimize` in dry-run mode to generate a diff of proposed changes
- 3User confirms and runs `python cli.py optimize --apply` to write the config
- 4Runs `python cli.py setup-heartbeat --provider ollama --apply` to route heartbeats to a local model
- 5Runs `python cli.py verify` to confirm the new config and view projected savings
OpenClaw config updated with Haiku as default model, Ollama handling heartbeats, 8KB session context, and prompt caching enabled, with a backup created and a savings report generated