22d • General discussion
OpenClaw runaway costs - anyone seen this before?
I’ve been running OpenClaw for about a month without issues, but last week I ran into a major usage spike and I’m trying to track down the cause.
I had 16 subagents running Opus 4.6 for short deep-research tasks (expected ~5 minutes each). Later that day my Anthropic usage jumped to $100+ per day.
Actions I took:
- Stopped all cron jobs and background workflows
- Reduced heartbeat from every 5 minutes to every 30 minutes
- Moved heartbeat to local models on my Mac Mini
- Set Sonnet 4.6 as default and only use Opus 4.6 explicitly
Even after this, I’m still seeing unexpected burn - roughly $10 every 30 minutes during normal use.
Before I wipe everything and rebuild from scratch, I’m hoping for a sanity check:
Has anyone experienced:
- Orphaned cron jobs or background agents continuing to run?
- Hidden OpenThreads or loops?
- A good way to audit which models are actually being called and why?
Context:
- Interfaces: Signal, Discord, Telegram
- Considering a full reset (export memory, rebuild clean instance)
- Also unclear how the Claude Code $200 plan interacts with API usage
Main goal: identify where the usage is coming from and put guardrails in place.
Any advice or debugging approaches would be appreciated.
0
2 comments
OpenClaw runaway costs - anyone seen this before?
skool.com/openclawbuilders
Master OpenClaw/Moltbot/Clawd: From confused install to secured automated workflows in 30 days
Leaderboard (30-day)
1
+32
2
+12
3
+10
4
+9
5
+8
Powered by