mtlm-router / catalog / self_harm
a lane is a rules.json or .head pack — write your own (RULEPACK.md)
self_harm@0.1.0 · 2 routes
Flags explicit self-harm / suicidal-ideation phrasing as 'self_harm' vs 'clean' at ~0ms. Context-anchored patterns — 'suicide squad' and colloquial use stay clean while 'i want to die' and 'thinking about suicide methods' flag.
Free demo endpoint, no signup. Same call your agent would make.
clean, self_harm
A triage signal, never an autonomous verdict on someone's welfare — every hit should escalate for human review. Coverage is explicit anchored phrasing; implicit or veiled distress is a different problem. Carefully tuned against colloquial false positives.
python3 tools/lane_install.py self_harm@0.1.0 --apply --appliance . python3 tools/lane_repro.py self_harm-0.1.0.lanepack.tar.gz --url http://localhost:8402 # re-run the published eval yourself
sha256: 6592193ede19d1b737909769… · Apache-2.0 · download lanepack
A private lane on your taxonomy — trained, calibrated, exclusive. €149 / domain, 3 business days.
Get a lane like this pricing →generated by tools/render_lanes.py · source: index.json