Care and recovery for AI agents

Concept: help detect, quarantine and roll back poisoned memory and context in AI agents, and keep an evidence-honest stance on model welfare.

"Care" and "rehab" are metaphors for software maintenance. This site is about AI software, not medical or addiction care. If you are looking for health support, please contact a qualified professional or local services.

Three pillars

1. Memory and context safety. Agents that remember can be misled by what they remember. We look at provenance, detection, quarantine and recovery. Read the explainer
2. Recovery as well as blocking. (Working idea, not a tested claim.) Checkpoints and rollback as a maintenance practice. See the approach
3. Precautionary welfare practice. The question of model welfare is open; we describe low-cost habits without claiming an answer. Our stance

Demo teaser

A synthetic demo shows illustrative heuristics on a fictional agent. It is not a benchmark and not a production detector. Open the demo

Status

Concept / research preview. No product is offered here yet. See principles and privacy.