#my openclaw has been AWEFUL! Can i fire it and get a new one?

5 messages · Page 1 of 1 (latest)

hollow jewel
#

suggestions racked up cost on Anthropic even after i told it to switch to gemini. it will cancel inexpensive jobs for cost and rack up cost iwht boated heartbeats and my telegram fails eveyday and i have to aks it to fix itself. how do i get telegram NOT to break? any ideas?

patent shoalBOT
#

Yeah, that sounds miserable. A few likely causes:

  1. /model only changes the current chat session
    It does not automatically retarget:
    • heartbeats (agents.defaults.heartbeat.model)
    • isolated cron jobs (openclaw cron edit <jobId> --model ...)
    • your global default (agents.defaults.model.primary)

So if you told chat “use Gemini”, heartbeats/jobs may still be burning Anthropic.

  1. Heartbeats are probably your biggest cost leak
    If you want them off:

    openclaw config set agents.defaults.heartbeat.every "0m"

    If you want them cheaper instead:

# 
   {
     agents: {
       defaults: {
         heartbeat: {
           every: "2h",
           model: "google/gemini-2.5-flash",
           isolatedSession: true,
           lightContext: true,
           target: "none"
         }
       }
     }
   }

isolatedSession + lightContext is the big token saver.

#

  1. Telegram breaking daily is usually transport/network
    First checks:

    openclaw status --all
openclaw channels status --probe
openclaw logs --follow

    If logs show fetch failed or getUpdates failed, common fixes are:

    {
  channels: {
    telegram: {
      network: { autoSelectFamily: false }
      // or proxy: "socks5://user:pass@host:1080"
    }
# 
     }
   }

Broken IPv6 / flaky VPS egress is a very common Telegram failure mode.

  1. Check cron jobs explicitly
    openclaw cron list
    If jobs were created with Anthropic pinned, they stay pinned until edited.

If you paste these, I can help narrow it down fast:

openclaw --version