🛡️ Privacy, security & compliance
Untrusted-email triage sandbox ★ Live
Inbound mail triaged by a throw-away LLM agent in an isolated VM, behind five defense layers.
Inbound email — the most hostile input there is — is triaged by a disposable, per-email LLM agent inside a dedicated VM, behind five defense layers (no-LLM ingest → injection-guard + PII-anon + quarantine → per-tenant blacklist → isolated VM with whitelisted tools only, no shell, no outbound SMTP → cross-tenant guard). The host never runs an LLM on raw mail; on anomaly the VM snapshot-reverts to clean. Live in production since 2026-05-14.