Computational Humor

Your agent jokes. You laugh — it remembers. Humor from embedding geometry, calibrated by your reactions.

Most agents are funny by accident. This one is funny on purpose, and gets funnier.

When the LIMBIC layer notices two semantically distant concepts that bridge in embedding space — the bisociation behind every good joke — it considers attempting humor. A sensitivity threshold (default 0.8) filters misfires. The next user message is scanned for laughter, emoji, affirmations, or dead air; that signal is captured and fed back into calibration so future attempts lean toward what actually lands. Frequency knobs go from off (zero overhead) to high (50% turn rate). Persona-aware humor preferences are read from Identity Persistence at startup if it's installed.

Install

openclaw plugins install @globalcaos/openclaw-computational-humor

Enable it in openclaw.json:

"plugins": {
  "allow": ["tinkerclaw-computational-humor"],
  "entries": { "tinkerclaw-computational-humor": { "enabled": true } }
}

Pairs Well With

• @globalcaos/openclaw-identity-persistence — your SOUL.md tells LIMBIC what flavor of humor matches the agent's persona, instead of generic LLM banter.
• @globalcaos/openclaw-total-recall — past laughter survives compaction. Calibration carries across sessions, not just within one.
• @globalcaos/openclaw-fractal-reflection — the agent reflects on what landed and what bombed; humor improves with the same loop that improves everything else.

👉 https://github.com/globalcaos/tinkerclaw 👉 https://thetinkerzone.com

Clone it. Fork it. Break it. Make it yours.