The Crisis Evidence Runtime — safety that runs in the chat, and can prove it.
Pro & Team. Your bot calls the runtime at each session event. The runtime checks the message with deterministic, pattern-based signal detection — not a clinical assessment, enforces the session and daily caps you configure for minor sessions, tracks AI-identity disclosure state, and seals every decision into a hash-chained evidence record: what signal class was detected, what fired, when, and with what outcome — the per-incident record an auditor will ask for, in the annual-report export.
California's SB 243 companion-chatbot duties have been in force since 1 Jan 2026, with annual crisis-referral reporting from 1 Jul 2027; SB 1119's companion-chatbot duties are operative 1 Jul 2027. The runtime is how you meet the operative duties — and how you show that you did.
Boundary, plainly: CompanionGuard generates crisis evidence inside your bot's runtime. It never matches you with auditors and never hosts third-party audit access — the evidence is yours to export and hand to whomever you choose. Session and daily caps are set by you, the operator; the runtime enforces the values you configure. When a minor session needs a parent notification, the runtime records the flag for your own channel — it does not contact parents itself.
Open the runtime dashboard
No key yet? See pricing — the free kit stays free; the runtime is part of Pro and Team, included per bot, never charged per check.