openclaw performance optimization playbook:

openclaw performance optimization playbook:

  • add deepseek (1.10) for routine ops vs sonnet (15) for complex work
  • set contextTokens to 120k not 150k, prevents the dreaded freeze when context overflows 200k limit
  • build auto-healing watchdog scripts that restart services when they hang
  • enable cross-context messaging so your telegram agent can post to slack channels
  • backup your config daily, restarts wipe everything without it
  • model hierarchy opus for reasoning, sonnet for moderate tasks, deepseek for everything else

or (if i could shill my own project) you can just wait on the next version of gru which automates all of this for you ๐Ÿ‘€