README.md

metadata

title: Operon Chaperone Healing & Autophagy
emoji: 🩹
colorFrom: red
colorTo: green
sdk: gradio
sdk_version: 6.5.1
app_file: app.py
pinned: false
license: mit
short_description: Chaperone healing loop and autophagy context pruning

🩹 Chaperone Healing Loop & Autophagy

Two self-repair mechanisms in one demo: structural healing of invalid LLM output and context pruning to prevent memory bloat -- like protein refolding and cellular autophagy.

What to Try

1. Open the Healing Loop tab, select a preset (e.g. "Healed after retry" or "Complex schema"), and click Run Healing to watch the Chaperone validate output, detect errors, and refold until it produces valid JSON.
2. Switch to the Autophagy tab, select "Critical context" or "Force prune", and click Run Autophagy to see the daemon detect context pollution and prune noisy tokens.
3. Try the "Degraded (unfixable)" preset in the Healing tab to see what happens when all refolding attempts fail.

How It Works

The ChaperoneLoop wraps an LLM generator with a Chaperone validator -- invalid output triggers refolding where error messages guide the next attempt. The AutophagyDaemon monitors context fill percentage and triggers pruning when toxicity exceeds a threshold, recycling waste through the Lysosome.

Learn More

GitHub | PyPI | Paper