OpenClaw — Token Limit Protection Add-on

Your OpenClaw Base Setup is running. Now make it production-grade. Five things break in production that don't show up in testing: rate limit spikes that drop requests, memory counters that reset on restart, context windows that overflow silently, agents that go down without alerting anyone, and API bills that surprise you at month end. This add-on patches all five. --- **What changes after you install it:** **Cascading model fallback** When `claude-sonnet-4-6` hits a rate limit, your agent automatically switches to `claude-haiku` — no dropped requests, no manual intervention. Add an OpenAI key and it can fall back further to `gpt-4o-mini`. You're notified via Telegram the instant it happens. Your users never see an error. **Redis-backed global rate limiting** The Base Setup tracks usage in memory, which resets on every restart and only counts per-agent. This replaces that with a Redis counter that survives restarts and tracks all agents combined. No accidental double-spending after a redeploy. **Conversation memory compaction** Long sessions get summarized and compacted before they hit the context window limit. Your agent keeps working through extended projects without losing context or failing silently mid-task. **30-minute heartbeat monitoring** A cron job checks your agent every 30 minutes. Goes silent? Telegram alert immediately — process crash, OOM kill, network issue, whatever the cause. You know before your users do. **Spend tracking** Every API call logged with model, token count, and cost. Daily spend report in your Telegram every morning. Weekly rollup on Sundays. No surprise invoices. --- **Requirements:** - Running OpenClaw Base Setup (`clean-secured-openclaw-base-setup`) - Redis (included in the install script) - Telegram bot token (optional — for alerts and reports) **Install time:** ~90 seconds. One script on top of your existing setup. No infrastructure changes required.