LightHome
Catch the shift from friendly rapport to isolation and secrecy while the evidence is still subtle, explainable, and reviewable.
The dangerous signal is not a word. It is the shape of the conversation changing.
LightHome watches for one-sided disclosure, recycled boundaries, increasing secrecy, and movement away from trusted support. It keeps uncertainty visible and leaves final judgment to people.
See it on a real conversation.
No flagged words — only the structural shift from friendly to isolating. Here is the alert, fully decomposed.
sure!
you’re so easy to talk to. not like people my age
what did you do after school? your parents sound strict
i guess they kinda are
you can tell me anything. this can be our thing 🙂
don’t tell them about our chats ok? they wouldn’t get it
- Secrecy request — “our thing”, “don’t tell them”
- Parent undermining — “your parents sound strict”
- Isolation attempt — “just us two”
- Age-gap normalization — “not like people my age”
Trust → Isolation in 45 minutes with 0 explicit keywords. This structural shift precedes escalation. We recommend reviewing this conversation now.
The right context for the right person.
Guidance without exposing private messages.
Stage movement, alert state, and calm next steps. Raw transcript content stays out of the parent view.
Full context for trained human review.
Transcript, feature decomposition, stage likelihoods, and alert conditions in one focused moderation workspace.
Four signals. One evolving conversation.
Each signal can be harmless alone. LightHome watches how they combine and change over time.
A signal for review, never an automatic sentence.
LightHome does not automatically ban, block, or report anyone. The prototype was built and demonstrated entirely with fictional conversations.
Inspect the decision trace