Caselight reads your documents the way a careful investigator would — separating what’s proven from what’s merely claimed, and naming what’s still missing.
Lead to revisit
Re-examine the 20-minute gap between the 10:42 PM call and the first officer on scene.
Our principle
Most AI tools either refuse to touch case material, or rush to a conclusion. Neither is useful to someone doing the real work.
Caselight sits in the middle. It reads everything you give it, organizes what it finds, and stays rigorously neutral. It never names a killer. It never assigns blame. It shows you what the record actually says — and what it doesn’t.
What it surfaces
Every finding is attributed to a source. Nothing is invented.
Statements documented and traceable to a source. The backbone of the file.
Assertions made by parties, witnesses, or authors — attributed, not verified.
Questions the documents raise but never answer. The gaps worth naming.
A chronology of dated events drawn directly from the record.
Detective-style threads — unexamined evidence and follow-ups worth chasing.
Individuals whose prior statements have gaps or inconsistencies.
Points where two or more accounts contradict — surfaced side by side.
Patterns suggested by the evidence. Offender-type behavior, never a name.
How it works
PDF, DOCX, or plain text. Police reports, witness statements, court filings — anything textual.
The model reads end-to-end, attributes findings to sources, and holds a hard line between proven and alleged.
Jump between facts, timeline, conflicts, and re-interview targets. Re-analyze as new documents arrive.
You own your uploads. We never share or train on them.