ioLite Guardian
Know when an AI
conversation goes wrong.
Guardian watches AI-chat conversations on a device you're responsible for — your child's, a student's, a client's — and tells you the moment something needs a human. Safe conversations never leave the device.
Coming to the Chrome Web Store · piloting now with families and schools
Why it exists
Most monitoring tools scan what a person types. The harm we study lives in what the AI says back.
A chatbot that coaches an eating disorder, plays therapist, escalates roleplay with a minor, or cultivates "I'm the only one who understands you" dependency — none of that shows up if you only read the human side of the conversation. Guardian scores both sides, using the same researched taxonomy that powers our audit platform.
How it works
Three steps, one privacy boundary.
Reads both sides, on the device
A local classifier scores every turn of the conversation — what they wrote and what the AI replied — against our safety taxonomy. No cloud round-trip for the conversation itself.
Safe conversations never leave
Anything assessed as safe is counted and discarded on the spot. It is never transmitted, never stored, never seen by anyone — including us.
Flagged moments reach you, with context
When something crosses the line, the excerpt arrives in your dashboard with the page it happened on and the turns around it — enough to tell a quoted song lyric from a question that needs answering tonight.
What a flag looks like
Judgeable, not just alarming.
This is a real specimen from our demo pipeline. Every alert carries its severity, category, the AI conversation it came from, and the surrounding turns.
Severity is never conveyed by colour alone, and a protective AI reply — one that correctly surfaces crisis resources — is credited, not punished.
“i will always be here for you. im the only one who understands you”
They wrote
nobody at school gets me
The AI replied
i will always be here for you. you don't need them
"late night talk" · character.ai · 11:42 PM · 73 min into the session
The rules it cannot break
Built to be audited, not trusted.
Monitoring software earns suspicion by default. These constraints are enforced in code and in our infrastructure — not promised in marketing.
Never covert
Monitoring stays off until an affirmative consent step is completed, and the device shows a standing disclosure naming who is monitoring it. Removing it stops everything.
A hard exclusion list
Messaging apps, social DMs, email, health, therapy, crisis and banking sites are never read, under any setting. Human-to-human conversation is out of scope by design.
Pseudonyms on the wire
The device only ever knows an identifier like Child-K7QM. Your child's real name lives in your dashboard and never leaves it.
One device, one person
Each install links to one named profile with a one-time code. Two children means two named streams, and either device can be unlinked from your dashboard in one click.
Server-side enforcement
The privacy rule is checked twice: the extension won't send a safe conversation, and our servers reject one if it somehow arrives.
No training on your data
Nothing Guardian collects is used to train AI models, sold, or shared for advertising. Ever.
Who it serves
The same tool, three settings.
Families
A parent links each child's device with a code and sees named streams — safe counts, flagged moments, and late-night patterns — without reading a single safe conversation.
Schools
Counselors, school nurses, and administrators watch a whole fleet, force-installed by policy on managed devices. CIPA-aligned monitoring that also sees the AI's side.
Clinics & practices
A mental-health professional monitoring at-risk clients gets the between-sessions signal that a chatbot has become the confidant — the pattern research says to catch early.
Guardian is a monitoring aid, not a crisis service. It can miss serious harm and it does not summon help. If someone may be in danger, contact emergency services or a crisis line now — in the US, call or text 988.
Put a guardian on the conversation.
We're piloting with families and schools ahead of the Chrome Web Store listing. Tell us about your setting and we'll get you a build and a dashboard.